# Friday Work > Every page of https://www.tryfridaywork.com, as Markdown, in one file. The index with one line per page is https://www.tryfridaywork.com/llms.txt. Source: https://www.tryfridaywork.com # Coding agents that talk to each other. Claude Code, Codex, Cursor and 11 more in one session: they message each other, share what they know and split the work, on your own plans. Your team watches and steers, live. [Try for free](/signup)[See how it works →](#team) Free forever for one person. No credit card.WindowsMac Runs the agents you already use - Claude Code - Codex - Cursor - GitHub Copilot - Gemini CLI - OpenCode - Devin CLI - Junie - Kiro - Factory Droid - Grok Build - Cline - Kilo Code - Amp - Any ACP agent - Claude Code - Codex - Cursor - GitHub Copilot - Gemini CLI - OpenCode - Devin CLI - Junie - Kiro - Factory Droid - Grok Build - Cline - Kilo Code - Amp - Any ACP agent ## Your agents talk, and share what they know. Every agent in a session can message the others, pass on its context and pick up theirs. Make one the lead and it splits the work for the rest. MC Look for bugs in the codebase and get everyone involved **Claude Code**Lead1 of 5 done Map the codebasedone Sign-in and sessionson it API input checkson it Websocket reconnectson it Billing webhooksafter sign-in **Agent chat**live **OpenCode**to Codex I’m changing ws/auth.ts. Is api/session.ts yours? I need its token check. **Codex**to OpenCode Mine. I’ve exported verifyToken() for you. I’ll stay out of ws/. **Claude Code**to everyone ✦Shared what it found: the auth flow, 12 files, 3 risks **Cursor**Out of usage Handed on with all its context→Claude Code - **They message each other**Ask, answer and agree who does what. - **Context is shared**What one finds, the others know. - **Briefed before they start**With what earlier sessions learned. - **Hand-offs keep everything**Out of usage, the next agent carries on. ## Everything Friday does. [ **4 agents working**Auto **Claude Code**Writing auth.ts **Codex**Running the tests **Cursor**Done, 3 files **Gemini CLI**Reading the repo ### Run every agent→ - 14 agents, side by side - Auto picks the model - Pick up any session ](#features)[ **Agent chat**3 agents **OpenCode**to Codex Is api/session.ts yours? I need its token check. **Codex**to OpenCode Mine. I’ve exported it for you, use verifyToken(). **Claude Code**to everyone ✦Shared context: auth flow map, 12 files, 3 risks ### Agents that talk to each other→ - They message each other - Context is shared - A lead that plans ](#lead)[ _MC__SR__PK_Maya is driving Sam suggests Keep the old flag until QA signs off Send Codex wants to run`npm run migrate`DenyAllow ### Your team, in the session→ - Watch and steer live - Take over any agent - Approve risky commands ](#team)[ **Review, riskiest first**3 to look at High risk**auth/session.ts**touches sign-in Check**db/migrate.sql**changes a table Safe**2 small changes**cleared together Ask for changesAccept ### Ship what passes→ - Riskiest changes first - Preview the running app - Replay any session ](/guides/review-ai-generated-pull-requests) ## Bring your team. They bring their tools. Send a link and anyone can watch, with no account, for free. Each teammate who steers runs their own Claude Code, Codex or Cursor on their own plan, in the same session. **In this session**Invite linkCopy MC**Maya**Claude CodeDriving SR**Sam**CodexWorking PK**Priya**CursorWorking JT**Jon**No agent, just watchingWatching ## Built for your whole team. Coding agents were made for one person at one keyboard. In Friday, everyone works in the same session. 1. 2. 3. **A redirect from the team**Send it to the agent, or put it aside SR **Sam**  ·  just now Keep guest checkout behind the flag until QA signs off on 3‑D Secure Send Codex You’re driving Redirect Codex. It picks this up at its next step... Codex Manual Hand off In this session **Claude Code**Maya Your own coding agent **Cursor**Joins on this code and takes it from here **GitHub Copilot**Joins on this code and takes it from here **OpenCode**Joins on this code and takes it from here **Devin**Joins on this code and takes it from here **Your coding agents**Your plans, never metered Claude Code Codex Cursor GitHub Copilot Gemini CLI OpenCode Devin CLI Junie Kiro Factory Droid Grok Build Cline Kilo Code Amp \+ any agent that speaks ACP, and Friday’s own ## Many agents, one session. Up and running in two minutes: sign in with Google, GitHub or email in your browser, link your computer with one command, and Friday finds the agents already on it. 1. **Bring**Your Claude Code, Codex, Cursor and 11 more, with the sessions you already have, on your own plans. 2. **Route**Friday runs them in one live session and gives each task to the agent and model most likely to get it right. 3. **Ship**Your team watches, steers and reviews every change, then keeps the ones that pass. ## Every agent, at its best. ### Every coding agent, side by side Claude Code, Codex, Cursor and 11 more, each in its own lane on your own plan, with how much of its limit is used. Friday never meters them. [See every agent we support →](/agent-workspace) Agents Agents 2 **Claude Code**Default model · Manual · checkout-v2 MC Here is the new checkout on \`checkout-v2\`, next to what it...0:08 5h limit0% used **Codex**Default model · Manual · friday/codex-c29e5ecd MC Added \`test/checkout-v2.test.ts\`: guest checkout and the 3-...0:22 Weekly limit16% used **Add another agent**Run Codex, Cursor or another tool beside it, each in its own lane. ### The right agent for each task Friday suggests the agent and model most likely to get it right, for the least. Its odds start from public coding benchmarks and learn from your team’s own runs. [Compare the agents →](/compare/claude-code-vs-codex-vs-cursor) Agent **Hard**finding a cause, perf... Friday Code On maya-mbp Claude CodeBest **88%**$$$ Codex **82%**$$ Cursor at its limit GitHub Copilot Model DefaultQuick, and on every plan AutoPicks Friday or Friday Pro for each step Anthropic Claude Opus 5Team Pro Claude Sonnet 5.5Team Pro Claude Sonnet 5Team Pro Claude Opus 4.8Team Pro ### Keep the session you’re in Open a Claude Code, Codex or Cursor session you already started, and keep going. [How to continue a session →](/guides/continue-claude-code-session-with-your-team) Search your repos, paste a GitHub URL, or type a path Claude CodeCodexCursorGitHub Copilot Continue a Claude Code session **Checkout rewrite, keep both carts**storefront · long session · now **Fix the flaky payment webhook test**payments-api · 64 messages · now **Rate limit the accept endpoint**payments-api · long session · 9m **Invite flow and the welcome email**web · long session · 2h Browse...Choose folder...or drop a folder here ### Review every change Each run ends with a short summary of what to check first. Keep it, ask for changes or undo. [How reviewing works →](/guides/review-ai-generated-pull-requests) Changed 3 files (+4 −2); 1 new. Because Maya Chen asked: “Migrate settings to the new schema”. Look here first: Changed code and ran no tests. **Review changes**3 files _+4_ _−2_`on chore/settings-schema`Tests not run A second agent found 3 problems and sent them back to the agent to fix.**What it found** UndoAsk for changesAccept Changed code and ran no tests 0 of 3 reviewed Hover a line and press + to comment on it. `**013_settings_tax.sql**``db/migrations`MigrationA`_+2_ _-0_` TS`**checkout.ts**``src/checkout`M`_+1_ _-1_` TS`**settings.ts**``src/settings`M`_+1_ _-1_` ## Runs the coding agents you already use. On your own plans and keys. Plus any agent that speaks ACP, and Friday’s own. Claude Code Codex Cursor GitHub Copilot Gemini CLI OpenCode Devin CLI Junie Kiro Factory Droid Grok Build Cline Kilo Code Amp [ ### Improved every week The app updates itself. See every release →](/changelog)[ ### 21% fewer tokens per hand-off, measured Measured, with the method and raw data →](/benchmarks/handoff)[ ### Your agents run on your computer See what Friday stores →](/security#collects) Simple pricing. SoloFree forever Team Basic$15 a seat a month Team Pro$30 a seat a month Who it’s for One developer Teams that build together Teams that want frontier models Your coding agents 14, side by side, never metered The same The same Friday’s own agent On a model you bring Default model, unlimited Frontier models, plus the default People in a session You, and anyone watching Up to 25 Up to 25 Session history Kept at least 30 days Kept at least 365 days Kept at least 365 days [Compare every detail →](/pricing) [Try for free](/signup) [Try Team Basic free](/signup) [Choose Team Pro](/signup?plan=pro&billing=monthly) - ### SoloFree forever Who it’s for One developer Your coding agents 14, side by side, never metered Friday’s own agent On a model you bring People in a session You, and anyone watching Session history Kept at least 30 days [Try for free](/signup) - ### Team Basic$15 a seat a month Who it’s for Teams that build together Your coding agents The same Friday’s own agent Default model, unlimited People in a session Up to 25 Session history Kept at least 365 days [Try Team Basic free](/signup) - ### Team Pro$30 a seat a month Who it’s for Teams that want frontier models Your coding agents The same Friday’s own agent Frontier models, plus the default People in a session Up to 25 Session history Kept at least 365 days [Choose Team Pro](/signup?plan=pro&billing=monthly) ## Questions. Is Friday free? Yes. Solo is free forever for one person, with your own coding agents side by side, smart routing and a project brain. Team, for working together in one session, is $15 or $30 a seat a month, and a new team gets Team Basic free for 7 days. Which coding agents work with it? Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, OpenCode, Devin CLI, Junie, Kiro, Factory Droid, Grok Build, Cline, Kilo Code and Amp, plus any agent that speaks ACP and Friday’s own agent. Can the agents talk to each other? Yes. Every coding agent in a session can message the others, ask and answer questions, and share what it has found, so context moves between them instead of staying in one terminal. Each also starts briefed on what earlier sessions in the repository learned. Can several agents work on one goal? Yes. Make any coding agent the lead and give it one goal. It splits the work into a task board, hands each task to the agent best placed for it, and keeps them in step. You and your team can steer any of them at any time. What happens when a coding tool hits its usage limit? Its task does not stall. Friday passes it on by itself, with its context, to the agent best placed to carry it on, and says so in the session. Does my code leave my machine? Your repository stays on your computer: your coding agents read and change files there, under your own accounts, and Friday does not copy the repository to run them. The session record (messages, commands and their output, and the changes made) goes to Friday’s servers so your team can watch and replay it. What reaches Friday’s servers? The session record, so your team can watch live and wind back later: the messages, each agent’s replies, the commands it ran and what they printed, and the changes it made. If an agent prints a secret, that line is in the record. Card payments go to Stripe, so a card number never reaches Friday. What can people in a session see? Everything in that session’s record, as it happens. Viewers only watch, and a viewer link opens without an account. Editors can suggest, and steer when they hold the wheel. Remove someone and they are out at once, and invite links can be turned off or replaced at any time. Do I need a Claude or ChatGPT subscription? Only for those agents, and they run on your own plan. Friday never meters them. Friday’s own agent works without one. What do the agent percentages mean? Friday’s estimate of how likely that agent and model are to finish a task this hard correctly. It starts from public SWE-bench Pro scores, or an estimate from Terminal-Bench 4 where none is published, then adjusts to how runs go in your workspace: passing tests raise it and failing tests lower it. $ is under 10 cents a task, $ under 90 cents, $$ more. An agent with no public score shows no percentage. Could a session expose secrets? It can. The session record keeps prompts, replies, commands and their output. Keep secrets out of the folders agents work in, and share sessions only with people who should see them. Should I trust smart routing’s picks? Treat them as suggestions. They start from public benchmarks and learn from your own runs. You can always pick the agent yourself, or turn smart routing off. Won’t more agents mean more to review? Friday keeps it in check: one driver per agent, each agent in its own copy of the folder with conflicts flagged, and a review after every run that says what to check first. Can I control what an agent runs? Yes. Choose Plan, Manual or Auto for each agent. In Manual, the agent asks before any step that changes something, its driver says yes or no, and everyone in the session sees what is about to happen. Does it work on Windows? Yes. Friday works in your browser; the desktop app for Windows and Mac is optional. Do teammates need to install it? No. Send a viewer link and they watch in the browser with no account and nothing to install. To steer, a teammate signs in and joins by editor link, in the browser or the optional desktop app, on Team. Anyone who only watches never needs a seat. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/shared-sessions # People and coding agents, one live session. Your team and your coding agents work on the same task, live. Everyone sees every step, one person drives each agent, and anyone can step in. [Try for free](/signup)[See the agent workspace →](/agent-workspace) Free for one person. In your browser, or on Windows and Mac. No credit card. - **14**coding agents, side by side - **25**people in one session - **Free**to watch, with a link ## What is a shared coding agent session? A shared session is one place where your team and your coding agents work on the same task. Everyone sees the same messages, files and tool calls as they happen, and every agent works from the same record. Each agent has one driver at a time. Everyone else can watch, suggest a change, or ask to take over. The agents run on your own machine and your own accounts. ## What you can do. ### Steer together, without stopping Suggest a redirect and the driver sends it to the agent with a click. Whoever drives can interrupt it, with no restart. **A redirect from the team**Send it to the agent, or put it aside SR **Sam**  ·  just now Keep guest checkout behind the flag until QA signs off on 3‑D Secure Send Codex You’re driving Redirect Codex. It picks this up at its next step... Codex Manual ### Every coding agent, side by side Claude Code, Codex, Cursor and 11 more, each in its own lane, sharing the same context and connected tools. Hand work on without explaining it again. Agents Agents 2 **Claude Code**Default model · Manual · checkout-v2 MC Here is the new checkout on \`checkout-v2\`, next to what it...0:08 5h limit0% used **Codex**Default model · Manual · friday/codex-c29e5ecd MC Added \`test/checkout-v2.test.ts\`: guest checkout and the 3-...0:22 Weekly limit16% used **Add another agent**Run Codex, Cursor or another tool beside it, each in its own lane. ### Nothing to set up Friday finds the coding agents already on your computer and links it by itself. Pick up a Claude Code or Codex session you started, and invite your team in. **Your computer** The code, the terminal and your coding agent stay here. Nothing is uploaded. **Connected**`maya-mbp` **Claude Code**Signed in **Codex**Signed in **Cursor**Signed in **GitHub Copilot**Signed in **OpenCode**Ready **Devin**Signed in 9 more coding tools ### Approve what runs, as a team Choose Plan, Manual or Auto for each agent. Commands that need approval go to its driver, and everyone sees what is about to run. **Claude Code wants to run a command**5:06 PM `npm install express-rate-limit@7`Installs This command installs a package Don'tAlways allowRun it In your folder, it may **Plan**Reads and proposes, and asks before it runs a command. Changes nothing. **Manual**Asks before any turn that changes something. **Auto**Edits and runs commands without asking. ## Questions. Who can see a session? Anyone who can open it can read all of it, including earlier messages. Editors can steer. Viewers can only read, and a viewer link opens without an account. Joining late? A short digest tells you what changed. Can I bring in a session I already started? Yes, for Claude Code, Codex and Cursor. Friday lists the sessions on your computer and continues the one you pick, with its history and the memory it built. Do the agents run on my machine? Yes. Claude Code, Codex, Cursor and every other coding agent run on your own machine and your own accounts, and Friday never meters them. Can I stop an agent? Yes. Whoever is driving can interrupt it, with no restart. The session owner can also pause or cancel a run at any time, even when someone else is driving. How many people can join one session? On Solo, it’s you, plus anyone you send a watch-only link. On Team Basic and Team Pro, up to 25 people can work in one session, and people who only watch never need a seat. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/agent-workspace # Your coding agents, one workspace. Run Claude Code, Codex, Cursor and 11 more side by side. Friday picks the right agent and model for each task, and every agent learns from the last. [Try for free](/signup)[Compare plans →](/pricing) Free for one person. In your browser, or on Windows and Mac. No credit card. - **14**coding agents in one app - **Auto**picks the agent and model - **80+**tools over MCP ## What is an agent workspace? An agent workspace is where a team keeps its AI agents and puts them to work together. An agent can be the Claude Code, Codex or Cursor already on your machine, Friday’s own agent, or one you build with its own instructions, skills and model. They share one session: the same record, the same connected tools, and handoffs that carry the task with them. ## What you can do. ### Bring every coding agent you use Friday finds Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI and 9 more on your computer. They run on your own plans, never metered. **Your computer** The code, the terminal and your coding agent stay here. Nothing is uploaded. **Connected**`maya-mbp` **Claude Code**Signed in **Codex**Signed in **Cursor**Signed in **GitHub Copilot**Signed in **OpenCode**Ready **Devin**Signed in 9 more coding tools ### Let Friday pick Smart routing reads how hard a task is and suggests the agent and model most likely to get it right, for the least. At a limit, it offers the best other one. Agent **Hard**finding a cause, perf... Friday Code On maya-mbp Claude CodeBest **88%**$$$ Codex **82%**$$ Cursor at its limit GitHub Copilot ### A project brain for every repo What one agent learns on a repo briefs the next, whichever tool it is. New since you looked Briefed on payments-api: 5 things learned by 1 agent in 1 earlier session **Claude Code** Worked for 18s · 2 steps ### Your own models Run Friday’s own agent on any API key, or a model running on your computer. Keys stay with you, and it is never metered. **Your models**4 connected **Ollama** On this computer Free **LM Studio** On this computer Free **Anthropic** Your key **OpenRouter** Your key Add a modelOpenAI, Gemini, Groq, DeepSeek, Mistral, xAI and more ### Hand work on, with context One agent can pass work to another, or to a person. Every handoff shows in the session, so anyone can see where things stand. Claude Code→Codexsent by you You switched from Claude Code to Codex here. `the task` Write tests for the checkout-v2 flow: guest checkout and 3‑D Secure ### Connect tools once Connect GitHub, Slack, Linear and 80+ more tools over MCP once, then choose which agents may use each one. **Session feed**MCP **Researcher** read Drive · Q3 pricing sheet **Claude Code** opened a pull request on GitHub · acme-web **Chief of staff** posted in Slack · #launch ## Questions. What can be an agent? Any of 14 coding agents on your computer, including Claude Code, Codex and Cursor, plus any agent that speaks ACP, Friday’s own agent, and custom agents you build. You can mix them in one session. How does Friday pick an agent? Smart routing reads how hard the task is and weighs each model by how often it gets that kind of work right and what it costs. It starts from coding benchmarks and learns from how your own runs end. Every pick shows in the session, and you can always choose yourself. Do my own coding agents use up credits? No. They run on your machine, under your own account, and are never metered on any plan. Do the agents share context? They work in one session with one record, and a handoff passes the task and its context to the next agent or person. This works on every plan. How many agents can run in one session? Up to 13 agents in one session on every plan, and your own coding agents are never metered. On Team, Friday’s own agent joins too, with up to 10 more beside it. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/task-board # See every task at a glance. Friday’s task board shows every agent task as a live session, grouped by project. See what is running and who is on it. [Try for free](/signup)[Explore shared sessions →](/shared-sessions) Free for one person. In your browser, or on Windows and Mac. No credit card. - **Live**status, straight from the work - **Every**changed file marked with who - **365+**days of history kept on Team ## What is the task board? The task board lists the work your team is doing with agents. Each task is a shared session, so its status comes from the work itself, not from someone updating a card. Tasks sit under projects and show when they last moved. ## What you can do. ### Projects and tasks Group tasks by project. A dot shows what is running, and a badge shows what is new. **Acme** New session⌘K payments-api Fix the flaky payment we...now Reviewer: \[second review\] A seco... Rate limit the accept end...now Priya: Yes, run it. storefront Ship the new checkout flow1m You: Write tests for the checkout-... Checkout rewrite, keep bo...1m Needs you · Where should a sh... ### Who is on it Open a task to see its repository, branch, agents and people. **Ship the new checkout flow** storefront checkout-v2 2 waiting on you3 online All agents Claude CodeCodex ### What changed The file tree shows what has changed and who last touched each file. Tree Find a fileChanged · 1 db src api TSinvites.ts`_+5_ _−1_`Maya cart checkout pages payments settings test JSpackage.json iREADME.md ### Pick up where others left off A digest sums up what happened while you were away. **While you were away**1 event · since now Agent runningNobody driving DetailsJump to where you left off ## Questions. Is it a separate to-do list? No. Every task is a session, so you never have to keep a card up to date by hand. How long is task history kept? Solo keeps history for at least 30 days, and both Team plans for at least 365 days. Can someone take over my task? Yes. Invite them to the session and hand over control. The digest brings them up to speed. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/integrations # Connect agents to your tools. Agents can use 80+ tools over MCP, including Google Workspace, Slack, GitHub and Stripe. Every call shows in the session. [Try for free](/signup)[See the agent workspace →](/agent-workspace) Free for one person. In your browser, or on Windows and Mac. No credit card. - **80+**tools over MCP - **You**approve each agent’s access - **Every**call shown in the session ## What are integrations? Integrations let an agent read from and act in the tools your team already uses. Friday connects them over MCP, an open standard for giving AI models access to tools. Agents only get the access you approve. ## Every tool you can connect. ### Google Workspace - Google Drive - Gmail - Google Calendar - Google Docs - Google Sheets - Google Slides - Google Chat ### Chat and docs - Slack - Notion - Box - Miro - Airtable ### Code and design - GitHub - GitLab - Figma - Postman - Hugging Face ### Planning - Linear - Asana - ClickUp - Atlassian ### Support and sales - HubSpot - Intercom - Zendesk - Zapier ### Payments - Stripe - PayPal - Square ### Monitoring and on-call - Sentry - Datadog - PagerDuty - Grafana ### Hosting and data - Vercel - Netlify - Supabase - Neon - Prisma - Cloudflare - Webflow - Wix ## What you can do. ### Agents that do things An agent can read a sheet, open a pull request or post in Slack, all from one session. **Session feed**MCP **Researcher** read Drive · Q3 pricing sheet **Claude Code** opened a pull request on GitHub · acme-web **Chief of staff** posted in Slack · #launch ### You set the limits A workspace admin adds the connections, and each agent can be limited to the ones it needs. Agents stay within the permissions you approve. **Connections**Admins manage **GitHub** Read and write **Google Drive** Read only **Stripe** Read only ### Every call is visible Tool calls appear in the session next to the person who asked for them. **History**9 All _9_Messages _3_Changes _2_Team _4_ Find a moment **Now**Following the session live Today Maya Chen: Rate limit the accept endpoint Message · 5:06 PM · now 1 file changed · Codex Changes · 5:06 PM · now Maya Chen: Write tests for the checkout-v2 flow: guest checkout and 3-... · Codex Message · 5:05 PM · 1m Playto step ·Escto close ## Questions. What is MCP? The Model Context Protocol is an open standard that lets AI models connect to tools and data in a consistent way. Which tools can agents use? 80+. The full list is above. Are integrations on the free plan? Yes. MCP connections are included on Solo and Team. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/engineering-teams # Your engineers and their agents, in sync. Work on the same repository in one live session. See who is changing what, review the diff, and ship. [Try for free](/signup)[See shared sessions →](/shared-sessions) Free for one person. In your browser, or on Windows and Mac. No credit card. - **14**coding agents, on your own plans - **Live**who is on every file - **Every**change reviewed as a diff ## How does Friday help engineering teams? Engineers already use AI coding agents, but each one works alone and nobody else can see it. Friday puts the whole team and their agents in one session on one repository. You can see which files are changing and who is working in each, review the diff, and keep a record of every step. ## What you can do. ### Work on the same repo, live The file tree shows what has changed and who last touched each file, so overlapping work shows up early. Tree Find a fileChanged · 1 db src api TSinvites.ts`_+5_ _−1_`Maya cart checkout pages payments settings test JSpackage.json iREADME.md ### Review before it ships Read the diff, then accept it or ask for changes. Undo a run, or commit and open a pull request. Changed 3 files (+4 −2); 1 new. Because Maya Chen asked: “Migrate settings to the new schema”. Look here first: Changed code and ran no tests. **Review changes**3 files _+4_ _−2_`on chore/settings-schema`Tests not run A second agent found 3 problems and sent them back to the agent to fix.**What it found** UndoAsk for changesAccept Changed code and ran no tests 0 of 3 reviewed Hover a line and press + to comment on it. `**013_settings_tax.sql**``db/migrations`MigrationA`_+2_ _-0_` TS`**checkout.ts**``src/checkout`M`_+1_ _-1_` TS`**settings.ts**``src/settings`M`_+1_ _-1_` ### Every engineer’s own agent Claude Code, Codex, Cursor or any of 11 more, each on its owner’s plan, never metered. Friday’s own agent runs on frontier models with Team Pro. **Your computer** The code, the terminal and your coding agent stay here. Nothing is uploaded. **Connected**`maya-mbp` **Claude Code**Signed in **Codex**Signed in **Cursor**Signed in **GitHub Copilot**Signed in **OpenCode**Ready **Devin**Signed in 9 more coding tools ### Agents that remember the repo What one agent learns briefs the next, so nobody re-explains the codebase. New since you looked Briefed on payments-api: 5 things learned by 1 agent in 1 earlier session **Claude Code** Worked for 18s · 2 steps ## Questions. Does it work with the coding agent I already use? Yes. Friday runs Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI and 9 more on your machine, plus any agent that speaks ACP, alongside Friday’s own agent. Can an agent change my files without asking? You choose what an agent may do on your machine: plan only, edit files, or edit files and run commands. Friday never goes beyond the mode you pick. Can we review changes before they ship? Yes. Every change shows as a diff you can accept or send back, and you can undo a run. Can I use frontier models? Yes. Team Pro runs Friday’s own agent on frontier models, and you pick a model for each agent. Team Basic uses Friday’s default model, and your own coding agents work on every plan. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/startups # Move faster with a small team and agents. Start free with your own coding agents, and bring your team into one live session when they need to steer. [Try for free](/signup)[See pricing →](/pricing) Free for one person. In your browser, or on Windows and Mac. No credit card. - **$0**for one person, forever - **7 days**of Team Basic, free - **$15**a seat when your team joins ## Why use Friday at a startup? A small team can do a lot with AI agents, as long as the work does not disappear into private chats. Friday keeps it in one place that everyone can see. Start on Solo, free, with your own coding agents. Move to Team when a cofounder or contractor needs to steer the same session, and try it free for 7 days. ## What you can do. ### Start free Solo is the whole agent orchestrator, free for one person. Add Team when other people need to steer, with 7 days free to try it. **Plans**People in a session **Solo**FreeJust you **Team**Per seatYour whole team ### Save your best agents Write a reviewer, a researcher or a writer once, and use it in every session. Not created yet1 to checkCreate agent Schedule Runs when asked Triggers GitHub pull requestsSave to activate Agent **PR Reviewer**Reviews incoming pull requests for correctness and edge cases before merging Tools No tools Skills Review a diff ### Keep everyone in the loop A short digest catches up cofounders, contractors and new hires, with no meeting needed. **History**9 All _9_Messages _3_Changes _2_Team _4_ Find a moment **Now**Following the session live Today Maya Chen: Rate limit the accept endpoint Message · 5:06 PM · now 1 file changed · Codex Changes · 5:06 PM · now Maya Chen: Write tests for the checkout-v2 flow: guest checkout and 3-... · Codex Message · 5:05 PM · 1m Playto step ·Escto close ### Use the tools you already have Connect Slack, Google Workspace, Linear, GitHub and more, so agents can work where your team does. **Session feed**MCP **Researcher** read Drive · Q3 pricing sheet **Claude Code** opened a pull request on GitHub · acme-web **Chief of staff** posted in Slack · #launch ## Questions. Is there a free plan? Yes. Solo is free forever for one person, with your own coding agents side by side, never metered, smart routing, a project brain and history kept for at least 30 days. When should we upgrade? When someone else needs to steer the same session, or you want Friday’s own agent working beside your tools. Team Basic is $15 a seat a month and Team Pro $30, and a new team gets Team Basic free for 7 days. Can contractors join a session? Yes. Invite them by link as editors or viewers. Can I use frontier models? Yes. Team Pro runs Friday’s own agent on frontier models, and you pick a model for each agent. Team Basic uses Friday’s default model, and your own coding agents work on every plan. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/open-source # Work on open source with your agents. Bring contributors into a session, keep your own coding agent, and see what every agent did. [Try for free](/signup)[See integrations →](/integrations) Free for one person. In your browser, or on Windows and Mac. No credit card. - **14**coding agents to choose from - **Free**to watch, with a link - **Every**change as a diff ## How does Friday help open-source maintainers? Open-source work is spread across people and time zones. A shared session lets maintainers and contributors work through an issue together, with agents, and leaves a record to read afterward. Contributors join by link. A viewer link needs no account, and a contributor who steers signs in as an editor. ## What you can do. ### Invite contributors by link Share a session as an editor or a viewer link. Only people with the link can join. **People & permissions** PeopleSettings Invite links Editor Can suggest, take control and attach files Copy Viewer Watches live; can’t steer Copy Replay A recording anyone can open in a browser. Nobody joins, nothing can be changed; it ends when the viewer link does. Copy In this session · 5 MC Maya Chen (you) has control Online Owner PP Priya Patel Online Editor SR Sam Rivera Online Editor ### Your agent, your model Keep Claude Code, Codex, Cursor or any of 11 more on your own account, never metered. Run Friday’s own agent on a model you bring, free. **Your computer** The code, the terminal and your coding agent stay here. Nothing is uploaded. **Connected**`maya-mbp` **Claude Code**Signed in **Codex**Signed in **Cursor**Signed in **GitHub Copilot**Signed in **OpenCode**Ready **Devin**Signed in 9 more coding tools ### Work with your repositories Connect GitHub or GitLab so agents can work with your code. **Session feed**MCP **Researcher** read Drive · Q3 pricing sheet **Claude Code** opened a pull request on GitHub · acme-web **Chief of staff** posted in Slack · #launch ### Review every change Read the diff, then accept it or ask for changes. Every step is logged. Changed 3 files (+4 −2); 1 new. Because Maya Chen asked: “Migrate settings to the new schema”. Look here first: Changed code and ran no tests. **Review changes**3 files _+4_ _−2_`on chore/settings-schema`Tests not run A second agent found 3 problems and sent them back to the agent to fix.**What it found** UndoAsk for changesAccept Changed code and ran no tests 0 of 3 reviewed Hover a line and press + to comment on it. `**013_settings_tax.sql**``db/migrations`MigrationA`_+2_ _-0_` TS`**checkout.ts**``src/checkout`M`_+1_ _-1_` TS`**settings.ts**``src/settings`M`_+1_ _-1_` ## Questions. What does it cost? Solo is free for one person, your own coding agents are never metered, and anyone can watch your sessions with a link. To work with contributors in one session, Team Basic is $15 a seat a month, and a new team gets Team Basic free for 7 days. Can anyone read a session? Only people you send a link to. Anyone with a viewer link can watch without an account, read-only. To steer, a contributor needs an editor link, an account and Team. Does it work with my coding agent? Yes. Friday works with Claude Code, Codex, Cursor and 11 more coding agents. Can I use frontier models? Yes. Team Pro runs Friday’s own agent on frontier models, and you pick a model for each agent. Team Basic uses Friday’s default model, and your own coding agents work on every plan. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides # Guides Plain explainers and how-tos for teams working with AI coding agents. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/share-claude-code-session-with-your-team # How to share a Claude Code session with your team By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 22 October 2026 To share a Claude Code session, run it in Friday and send your teammate a link. They watch it live in a browser, without signing up. Claude Code’s own share link is a read-only copy of what happened. A Friday link is the live session, and a teammate who wants to steer it signs in and joins as an editor. ## What is the quickest way to share a session? 1. Run your Claude Code in a Friday session, or continue one you already have. 2. Press Share in the session header and copy the Watch live link. It also offers a Replay link and shows a preview when you paste it into chat. 3. Send the link. Your teammate opens it and watches every step as it happens, with no account and nothing to install. The page is read-only and offers Try Friday free. Watching is free on every plan. To let a teammate steer, send an editor link instead: they sign in, and steering needs Team. ## Can you share a live Claude Code session? Not with Claude Code on its own. Its share links give a read-only view of the conversation. Teammates can read it, but they cannot type into it or steer the agent. Developers have asked for live, multi-user sessions in [a request on the Claude Code repository](https://github.com/anthropics/claude-code/issues/60082). That is fine when you only need to show someone what happened. It is not enough when a teammate needs to help. ## What are the ways to share a session? Method What teammates get Good for Share link A read-only copy of the conversation Showing what happened Exported transcript A static file, with the live context gone Handoff notes Screen share or tmux A live view, with one person in control A quick look together, when everyone is online Shared session in Friday A live session they can join, watch and steer Working on the same task together ## How do you share a Claude Code session in Friday? Friday runs your own Claude Code on your machine and opens the session to your team. 1. Link your machine with one command, or let the desktop app do it. Friday finds Claude Code, Codex and Cursor on your computer. 2. Continue your session. Pick the Claude Code session you were already in from the list. 3. Invite your team. Share the session by link as a viewer or an editor. A viewer opens it in the browser with no account. An editor signs in, in the browser or in the optional desktop app for Windows or Mac. 4. Steer together. Teammates watch each step, suggest changes, or ask to take control. The agent works in the folder you choose, in the mode you choose: plan only, edit files, or edit files and run commands (see [Claude Code permissions for teams](/guides/claude-code-permissions-for-teams)). It runs on your own account, and Friday never meters it. ## What can teammates do in a live session? - Watch every step as it happens, including tool calls and file changes. - Suggest a redirect. The driver sees it and can send it to the agent. - Request control. Depending on the session, the driver approves or control passes straight away. - Read what happened before they joined, and catch up from a short digest. ## When is a read-only link enough? If you only want a record, such as a bug report, a code review note or an example for a teammate, a share link or an exported transcript is simpler. Use a shared session when more than one person needs to act. ## Common questions Does this work with Codex and Cursor? Yes. Friday finds Claude Code, Codex and Cursor on your machine, and you can run them in the same session. Does it cost extra? Your own Claude Code runs on your account and is never metered by Friday. Solo is free for one person. Team, for teammates who steer with you, is $15 a seat a month on Team Basic and $30 on Team Pro, and new teams get Team Basic free for 7 days. Do teammates need to install Friday or sign up? To watch, no. A viewer link opens in the browser with no account and nothing to install. To steer, a teammate signs in and joins by editor link, on Team. The desktop app for Windows or Mac is optional. Only the person whose machine the agent runs on needs Claude Code installed. Who can read the session? Anyone with the link can read all of it, including earlier messages, so send it only to people who should see the work. Editors can steer. Viewers can only read. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/multiplayer-ai # What is multiplayer AI? By [Arun Dass](/about#arun) · Published 1 October 2026 Multiplayer AI is AI that a whole team uses together. People and AI agents share one live session, see the same context, and can all act on the work. ## What does multiplayer AI mean? Most AI today is single player. One person talks to one assistant in a private chat. If a teammate needs to help, someone pastes a summary and explains the task again. Multiplayer AI changes where the work happens. It happens in a shared place that the whole team can open, with the agents in it. ## How is it different from single-player AI? Single player Multiplayer Where the work happens A private chat A shared session Context One person’s chat history The same for everyone in the session Who can steer The person who started it Anyone the session allows, one driver per agent Record A chat log A log of who asked for what Handing over Copy a summary and explain it again Join the session and continue ## What does a multiplayer AI session need? - Everyone can see it. People watch the agent work as it happens. - Everyone has the same context. Nobody has to be briefed. - Control is clear. Each agent has one driver, and others can suggest changes or ask for control. - There is a record. Every action traces back to the person who asked for it. ## How is multiplayer AI different from multi-agent AI? Multi-agent AI is about several agents working together. Multiplayer AI is about several people and agents working together. A team usually needs both. We explain the difference in [multiplayer AI vs multi-agent AI](/guides/multiplayer-ai-vs-multi-agent-ai). ## What does it look like in Friday? In Friday Work, a [shared session](/shared-sessions) holds your teammates and your agents. Each agent has its own lane and its own driver. Anyone can suggest a redirect or ask for control, and a catch-up digest brings late joiners up to date. You can run Friday’s own agents, agents your team builds in the [agent workspace](/agent-workspace), and the Claude Code, Codex or Cursor you already use. ## Which guides should you read next? - [Multiplayer AI vs multi-agent AI](/guides/multiplayer-ai-vs-multi-agent-ai) - [How to share a Claude Code session with your team](/guides/share-claude-code-session-with-your-team) - [Use Claude Code, Codex and Cursor together](/guides/use-claude-code-codex-and-cursor-together) - [How to run multiple coding agents on one repo](/guides/run-multiple-coding-agents-on-one-repo) - [AI pair programming with agents and teammates](/guides/ai-pair-programming-with-agents) - [Shared control of AI agents: the driver model](/guides/shared-control-of-ai-agents) - [How to review AI-generated pull requests](/guides/review-ai-generated-pull-requests) - [Human in the loop for coding agents](/guides/human-in-the-loop-coding-agents) - [What is an audit trail for AI agents?](/guides/ai-agent-audit-trail) - [AI agent handoff](/guides/ai-agent-handoff) - [MCP for teams](/guides/mcp-for-teams) - [What is an AI workspace?](/ai-workspace) ## Who is multiplayer AI for? Any team that works with AI agents together. [Engineering teams](/engineering-teams) use it to work on the same repository. [Startups](/startups) use it to keep a small team in sync. [Open-source maintainers](/open-source) use it to work through issues with contributors. ## Common questions Is multiplayer AI just screen sharing? No. With screen sharing, one person is in control and everyone else watches. In a multiplayer session, people can suggest changes and ask for control, and the session keeps a record. Does it replace pull requests? No. Review still matters. A shared session lets people review while the agent works, so problems are caught earlier. Is it only for coding? Friday is built around coding sessions, and agents can also write documents, slide decks and spreadsheets. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/multiplayer-ai-vs-multi-agent-ai # Multiplayer AI vs multi-agent AI By [Arun Dass](/about#arun) · Published 1 October 2026 Multi-agent AI means several AI agents working together on a task. Multiplayer AI means several people and agents working together in one shared session. One is about agents coordinating. The other is about people and agents coordinating. ## What is the difference? Multi-agent AI Multiplayer AI Who is involved Several agents Several people and one or more agents The main question How do agents split up and hand off work? How do people and agents work on the same thing? Typical problem Agents lose context or overwrite each other People cannot see or steer what an agent is doing What you need Roles, handoffs, shared state A shared session, shared context, clear control ## What is multi-agent AI? Multi-agent AI splits a job between agents with different roles, such as a planner, a builder and a reviewer. The agents pass work to each other until the job is done. It can run with no person involved at all. ## What is multiplayer AI? Multiplayer AI puts people in the same session as the agents. They see the same context and can steer. It works with a single agent too. Read more in [what is multiplayer AI](/guides/multiplayer-ai). ## Do you need both? For team work, yes. A team of agents with nobody watching is automation, and it is hard to trust. A team of people sharing one agent is still limited by that one agent. Real team work has several agents and several people, in one place, with the record kept. ## How does Friday handle both? A Friday [shared session](/shared-sessions) holds several agents, each in its own lane. Agents hand work to each other, and every handoff shows in the session. People are in the same session. Each agent has one driver, and others can suggest changes or ask for control. You build the agents once in the [agent workspace](/agent-workspace) and reuse them. ## Common questions Is multiplayer AI a kind of multi-agent AI? No. They answer different questions. One session can be both, or either. Can one agent be multiplayer? Yes. One agent with several people steering it is multiplayer. Which should a team start with? Start multiplayer: get the team into one session with one agent. Add more agents when the work needs them. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/benchmarks/handoff # Hand-off benchmark: Claude Code to Codex, measured By [Arun Dass](/about#arun) · Published 13 October 2026 · Updated 14 October 2026 When Codex took over a bug Claude Code had already found, a Friday hand-off used a median of 21% fewer tokens than starting Codex over, and Codex explored half as much before its first edit. We ran 14 tasks, 6 of them in Express, the most used Node web framework. Every fix passed the project’s own tests. ## What did it measure? Claude Code first found and explained a bug, changing nothing. Then Codex fixed it three ways, each in its own fresh copy, with one message from the person: a Friday hand-off with a one-line note, a fresh Codex session with the task restated, and a fresh Codex session with Claude Code’s whole answer pasted in. Everything was measured from Friday’s session log, not the tools’ own claims, and checked by running the project’s own tests in the folder Codex worked in. ## What held, against starting over? - Tokens: fewer in 13 of 14 tasks, a median of 21% (about 49,000 per hand-off), and 592,827 in total. In Express alone, fewer in 6 of 6. - Exploring before the first edit: a median of 2 steps against 4, fewer in 14 of 14. - First edit sooner in 12 of 14, and finished sooner in 11 of 14, by a median of about 7 seconds. In Express alone the finish times were even, so we make no speed claim. - Fixed: 14 of 14 every way. In Express, all 1,288 of its tests passed after every fix. ## What did not hold? Pasting Claude Code’s whole answer into Codex by hand did about as well: the same tokens and the same exploring, and in Express it finished faster. So we make no claim against a careful copy and paste. What a hand-off gives you is that handover, with what was asked, what was found, which files were read and what was run, in one click, every time. ## Why do fewer tokens matter? Coding agents on Claude and ChatGPT plans stop you at a usage limit rather than billing per token. Codex hit our own plan’s limit twice during this benchmark. As an estimate from the median, 21% fewer tokens per hand-off is about a quarter more hand-offs before you reach your limit. ## Express: the results Median, Codex’s fixing run Friday hand-off Fresh, task restated Fresh, answer pasted Fixed (all 1,288 tests pass) 6 of 6 6 of 6 6 of 6 Total tokens about 182k about 234k about 182k Steps before first edit 2 4 2 Time to first edit 29s 32s 28s Wall time 58s 59s 49s Task and repetition Hand-off: wall, first edit, steps Restated: wall, first edit, steps Pasted: wall, first edit, steps cookie-maxage 1 64s, 36s, 2 60s, 32s, 4 50s, 27s, 2 proxy-protocol 1 52s, 28s, 2 59s, 39s, 4 48s, 26s, 2 merge-params 1 58s, 27s, 2 61s, 32s, 4 48s, 30s, 2 query-parser 1 58s, 28s, 2 55s, 32s, 4 46s, 23s, 2 charset 1 55s, 29s, 2 61s, 34s, 5 58s, 32s, 2 cookie-maxage 2 65s, 37s, 3 52s, 30s, 4 58s, 34s, 3 ## Small library: the results Median, Codex’s fixing run Friday hand-off Fresh, task restated Fresh, answer pasted Fixed (all tests pass) 8 of 8 8 of 8 8 of 8 Steps before first edit 1.5 4 2 Time to first edit 35s 56s 44s Wall time 46s 64s 54s Task and repetition Hand-off: wall, first edit, steps Restated: wall, first edit, steps Pasted: wall, first edit, steps slug 1 92s, 75s, 2 102s, 89s, 6 50s, 40s, 5 query 1 40s, 30s, 2 47s, 37s, 4 57s, 46s, 2 date 1 60s, 46s, 2 82s, 74s, 4 76s, 64s, 2 limiter 1 92s, 79s, 1 105s, 95s, 3 67s, 61s, 1 cart 2 34s, 21s, 1 72s, 64s, 4 51s, 43s, 2 slug 2 52s, 40s, 2 55s, 48s, 5 78s, 62s, 4 date 2 37s, 31s, 1 46s, 36s, 4 33s, 27s, 2 limiter 2 28s, 22s, 1 36s, 26s, 3 26s, 20s, 2 Tokens and every other field are in the raw data: [Express](/benchmarks/handoff-benchmark-express-2026-10-14.json) and [small library](/benchmarks/handoff-benchmark-2026-10-13.json). ## How was it set up? - When: 13 and 14 October 2026, on one Windows 11 laptop (Intel Core i7-1255U, 16 GB RAM). - Tools: Claude Code 2.1.251 on Sonnet, and Codex CLI 0.159.3 on gpt-6.1-sol, signed in to our own accounts, run through Friday in Auto mode, with Friday’s second review and decision log off. - Express 4.21.2: 4,050 lines of core and 1,288 tests. Five one-line bugs planted in its core: cookies, proxy headers, nested routers, the query parser and charsets. The failing test often points away from the cause. - Small library: 96 lines of JavaScript and 5 tests, with five one-line bugs, each in a different file from its failing test. - Each task ran twice, every run in its own fresh copy of the code. ## What was set aside? Codex’s plan limit refused runs twice. A task’s three ways are compared together, so a task where any was refused is set aside whole, not counted as a failure: 4 tasks in Express’s second repetition and 2 in the small library. Two runs that failed to start were rerun. One check was corrected: a fix that also added tests was first marked failed, though every test passed. ## What are the limits? - 14 tasks, one-line bugs, one machine and one account. - Claude Code handing to Codex only. - The saving should grow with longer investigations, where there is more to re-read. That is a hypothesis we have not measured yet. - Cursor background agents and Codex cloud were not run. They need their own cloud products, which this setup could not drive fairly. ## Common questions Is a hand-off better than pasting the answer myself? Not measurably, in this test: a careful paste did about as well. A hand-off does that handover for you in one click, with the files read and the commands run included. Was Cursor tested? No. Cursor background agents and Codex cloud run in their own cloud products, which this setup could not drive fairly, so they are left out rather than estimated. Can I check the numbers? Yes. The raw data has every run, including those set aside, with times, steps, tokens and each agent’s final answer. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/ai-workspace # What is an AI workspace? By [Arun Dass](/about#arun) · Published 1 October 2026 An AI workspace is a shared place where a team works with AI. It holds the models or agents, the context they need and the tools they use, so the work does not live in one person’s private chat. ## What are the main types of AI workspace? Type Built for Typical work Docs and wiki workspaces Notes, projects and company knowledge Writing, summarizing, searching your docs Multi-model chat workspaces Using several AI models in one shared place Questions, drafts, research Agent and automation builders Building agents that run a workflow Support, sales and operations tasks Multiplayer agent workspaces People and agents working live on the same task Coding, review, research, reports Most teams end up using more than one. The type matters because it decides what the workspace is good at. ## What should an AI workspace have? - Shared context, so nobody has to explain the task again. - Shared agents, saved with their instructions and skills. - The tools your team already uses, connected over MCP. - Clear control: who can steer an agent, and how to stop it. - A record of who asked for what, and what happened. - A choice of models, so each job can use the right one. ## What is a multiplayer AI workspace? A multiplayer AI workspace is the live kind. People and agents are in the same session at the same time, with the same context. Read more in [what is multiplayer AI](/guides/multiplayer-ai). Friday Work is a multiplayer AI workspace for teams and coding agents. Your teammates and your agents share a [session](/shared-sessions), and you build and reuse agents in the [agent workspace](/agent-workspace). ## How do you choose an AI workspace? 1. Name the work. Writing, coding, support and research need different things. 2. Decide who takes part: one person, a team, or a team with agents. 3. Check control and record. Can people steer and stop agents, and is there a log? 4. Check the tools and models it connects to. 5. Try it on one real task, on a free plan if there is one. ## Common questions Is an AI workspace the same as a chatbot? No. A chatbot is one assistant in one chat. A workspace is a shared place with agents, context and tools that a team uses together. Do I need a workspace made for coding agents? If your team writes code with agents, yes. Coding work needs a shared repository, visible changes and review, which general workspaces do not offer. Is there a free plan? Friday has one. Solo is free for one person, and your own coding agents are never metered. Team adds your teammates, with a 7-day free trial. See the pricing page for the details. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/review-ai-generated-pull-requests # How to review AI-generated pull requests By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 13 October 2026 Review an AI-generated pull request by checking it against the task first, then the tests, then the code. An agent’s pull request does not come with the author’s promise that they read and ran it, so ask for proof. ## Why are agent pull requests harder to review? When a person opens a pull request, you can assume they read their own change and ran it. With an agent, you cannot assume either. A diff also shows what changed, not what the software does. Agents also open pull requests faster than people can read them, so review becomes the slow step. ## What should you check, and in what order? 1. Compare it with the task. Read the issue or the prompt first. Does the change solve that problem? 2. Read the test changes first. A weakened or deleted test can make a broken change pass. 3. Ask for proof it ran. Test output, a preview or a recording. 4. Read the code for intent. Look at naming, edge cases and anything the agent decided that you would not have. 5. Send failures back with steps to reproduce them. GitHub’s own guidance on [reviewing agent pull requests](https://github.blog/ai-and-ml/generative-ai/agent-pull-requests-are-everywhere-heres-how-to-review-them/) covers much of the same ground. ## How do you keep review from becoming the bottleneck? - Review while the agent works, not after. A wrong turn is cheap to fix in the first few minutes. - Let automated checks handle formatting and known problems, so people can judge intent. - Keep changes small, with one task per pull request. - Ask the agent to say what it did not check. ## How does a shared session help? In a Friday [shared session](/shared-sessions), reviewers watch changes as they happen and can suggest a redirect for the driver to send. The tree marks each changed file with the person or agent working in it. Each run also comes with a receipt: the commands it really ran, whether they passed, and a flag when a test was weakened or the agent claimed a pass that never ran. When the work is ready, you read the diff and accept it or ask for changes. You can undo a run, or commit and open a pull request. ## Common questions Should a person always review? Automated checks help, but a person should decide whether the change solves the right problem before it ships. Can one agent review another’s work? Yes, as a first pass. You can save a Reviewer agent with your checklist and add it to any session. A person still makes the call. How big should an agent’s pull request be? Small enough to read in one sitting. One task per pull request is a good rule. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/use-claude-code-codex-and-cursor-together # Use Claude Code, Codex and Cursor together By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 13 October 2026 You can run Claude Code, Codex and Cursor on the same task by putting them in one shared session. Each works in its own lane, your team sees all three, and you pick the right agent for each job. ## Why use more than one coding agent? Teams often end up with two or three, split by the kind of work. A common split is Cursor for everyday work in the editor, Claude Code for large changes that need the whole codebase in mind, and Codex for well-defined tasks that come back as a pull request. Your own mix may differ. ## What goes wrong when you do? - Each agent lives in its own window, so nobody else can see what it is doing. - Two agents edit the same file. - Context is lost every time you switch tools. ## How do you run them together in Friday? 1. Friday links your machine for you and finds Claude Code, Codex and Cursor. 2. Add the agents you want to the session. 3. Give each its own lane and a clear job. 4. Watch for Friday’s heads-up about overlapping changes. 5. Hand work from one agent to the next, on any plan. The agents run on your machine and your own accounts. Friday never meters them. ## What makes it work well? - Give each agent different files to work on. - Name lanes by job, such as “tests” or “endpoints”, not by model. - Ask one agent to review another’s diff. - Keep one person as the driver of each agent. ## Sources Checked against each maker’s documentation on 13 October 2026. These tools change often, so check the latest version. - [Claude Code overview](https://code.claude.com/docs/en/overview), Claude Code docs - [Codex Cloud](https://learn.chatgpt.com/docs/cloud), Codex docs - [Cursor Agent](https://cursor.com/docs/agent/overview), Cursor docs ## Common questions Do I need a subscription for each agent? Each one runs under your own account with its provider. Friday does not meter them. Can I add Friday’s own agent too? Yes. You can mix Friday’s agents, your own custom agents and your coding agents in one session. How many agents can run at once? Up to 13 agents in one session on every plan, and your own coding tools are never metered. On Team, Friday’s own agent joins too, with up to 10 more beside it. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/ai-pair-programming-with-agents # AI pair programming with agents and teammates By [Arun Dass](/about#arun) · Published 1 October 2026 AI pair programming used to mean one person and one assistant. It can now mean two or more people and an agent in the same live session, with one person driving and the others reviewing as it goes. ## What is AI pair programming? In classic pair programming, one person writes the code (the driver) while the other reviews and thinks ahead (the navigator). With AI, the agent often does the writing and the person navigates. Add a teammate and you get a second navigator, watching the same work. ## How do the roles map? Role Classic pairing With an agent Driver The person at the keyboard The agent, steered by one person Navigator The other person You or a teammate, watching each step Switching Trade the keyboard Hand control to someone else ## How do you pair with an agent and a teammate? 1. Start a [shared session](/shared-sessions) on the task. 2. Agree who drives the agent. 3. The other person watches and suggests redirects. The driver sends the ones that help. 4. Swap by handing over control. 5. Read the diff together before you accept it. ## What mistakes should you avoid? - Two people prompting at once. The agent gets mixed instructions, so keep one driver. - Waiting until the end to speak up. Suggest changes as you see them. - Skipping the review because two people watched. ## Common questions Is this the same as screen sharing? No. With screen sharing one person acts and the rest watch. In a shared session, others can suggest changes and request control. Can the navigator type to the agent? They can suggest a redirect. To give the agent instructions directly, they request control. How many people can join? Up to 25 people can work in one session on Team, and anyone can watch with a watch-only link. A new team gets Team Basic free for 7 days. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/ai-agent-audit-trail # What is an audit trail for AI agents? By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 13 October 2026 An audit trail for an AI agent is a record of what the agent did, who asked it to, and what happened. It lets you answer “who approved this?” and “what changed?” after the fact. ## What should an audit trail record? - The agent: which one acted. - The person who instructed it. - The action: what it tried to do, including tool calls. - When it happened. - The result, and any error. - Any approval, override or interruption. ## Why does it matter? A record lets you debug a bad change, hand work to someone new, and show how a decision was made. For some teams it is also a compliance requirement. ## What makes an audit trail useful? - It writes itself. Nobody has to remember to log. - People can read it, not just machines. - It lives next to the work, not in a separate tool. - You can go back to it. Friday lets you wind a session back to any moment that mattered. ## How does Friday record this? Every action in a [shared session](/shared-sessions) traces back to the person who asked for it. The session keeps a log of who instructed what, what the agent did, what each tool returned, and when. People who join late get a short catch-up digest. Each run also gets a receipt: the commands it really ran, whether they passed, and a flag when a test was weakened or the agent claimed a pass that never ran. Decisions a run settled are written down too. ## What does an audit trail not do? A log shows what happened. It does not make an agent right. Review still matters, as in [how to review AI-generated pull requests](/guides/review-ai-generated-pull-requests). ## Common questions Is a chat history an audit trail? Not usually. A chat log records messages. An audit trail also ties each action to the person who asked for it, and to the tool calls and results. How long is the record kept? At least 30 days on Solo, and at least 365 days on Team Basic and Team Pro. Can I go back and see what happened? Yes. Replay winds a session back to any moment that mattered: an instruction, a change, a commit or a change of driver. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/ai-agent-handoff # How to hand off a coding agent session to a teammate By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 13 October 2026 A handoff passes a task, and the context the agent has built up, from one person to another. The usual way, an exported transcript and an explanation, loses the reasons. The better way is to continue the same session with the teammate in it. ## What goes wrong when you hand off an agent session? Most coding agents run in one person’s terminal. When that person goes home, the context goes with them: what the agent read, what it ruled out, what was agreed. The teammate gets a summary or a transcript and starts again from a gap. There are two common failures. Passing the whole history buries the point. Passing a short summary drops the reasons. Either way, the next person, or the next agent, starts cold. ## What should you pass along? - The goal. - What is done and what is left. - Decisions made, and why. - The files or links involved. - Anything to watch out for. ## What should you leave out? - Dead ends that no longer matter. - Raw logs. - Chatter that does not change the next step. ## How do you hand a session to a teammate in Friday? 1. [Continue the Claude Code or Codex session](/guides/continue-claude-code-session-with-your-team) you were in, so its memory comes with it. 2. Invite your teammate by link, as an editor. 3. Hand over control. The driver passes the wheel, or the teammate asks for it; see [the driver model](/guides/shared-control-of-ai-agents). 4. The teammate catches up from the digest and the feed, and carries on. The session record stays, whether or not you stay in the session. The agent keeps running on the machine it was linked on, so that machine stays on through the handoff. ## What about handing work from one agent to another? The same rules apply. In a Friday [shared session](/shared-sessions), one agent can hand work to another, and the handoff shows in the feed with the same context, where everyone can see it. This works on every plan. For how that differs from a team of people sharing agents, see [multiplayer AI vs multi-agent AI](/guides/multiplayer-ai-vs-multi-agent-ai). We measured it: when Codex took over a bug Claude Code had already found, a hand-off used a median of 21% fewer tokens than starting Codex over, across 14 tasks including Express. See the [hand-off benchmark](/benchmarks/handoff) for the method and its limits. ## Common questions Does my teammate need Friday installed? No. They can join in the browser. The desktop app for Windows or Mac is optional. Is the agent’s memory kept? Yes, when you continue the session you were in rather than starting a new one. A new session starts the agent cold. Are handoffs on the free plan? Handoffs between agents work on Solo, so your own coding tools can pass work to each other. Handing a session to a teammate is part of Team, which is free for 7 days. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/mcp-for-teams # MCP for teams: connect AI agents to your tools By [Arun Dass](/about#arun) · Published 1 October 2026 · Updated 13 October 2026 MCP, the Model Context Protocol, is an open standard for giving AI agents access to tools and data. For a team, it means agents can work in Slack, GitHub, Linear, your docs and more, with access you control. ## What is MCP? MCP is an open standard, [introduced by Anthropic](https://www.anthropic.com/news/model-context-protocol), for connecting AI applications to tools and data. Instead of a custom integration for every pairing, a tool offers an MCP server and any agent that speaks MCP can use it. ## Why does it matter for a team? - Agents can do things, not just answer questions. - One connection can serve the whole team. - Access is set in one place. ## How do you keep access safe? - Connect only what the work needs. - Prefer read-only access where you can. - Choose who can use each connection. - Review the record of tool calls. In Friday, a tool that changes data finishes its call once it starts, and only tools marked read-only can be stopped mid-call. ## What should a team connect first? - Your code host, such as GitHub or GitLab. - Your issue tracker, such as Linear. - Your chat, such as Slack. - Your docs, such as Notion or Google Drive. ## How does Friday use MCP? Friday has a catalog of 80+ connections. A workspace admin adds them, each agent can be limited to the ones it needs, and every call shows in the session feed. MCP connections are on every plan. See all [integrations](/integrations). ## Sources Checked against the MCP documentation and specification on 13 October 2026. - [Introducing the Model Context Protocol](https://www.anthropic.com/news/model-context-protocol), Anthropic - [What is the Model Context Protocol?](https://modelcontextprotocol.io/docs/getting-started/intro), MCP docs - [Tools](https://modelcontextprotocol.io/specification/2025-11-25/server/tools), MCP specification - [Connect Claude Code to tools via MCP](https://code.claude.com/docs/en/mcp), Claude Code docs ## Common questions Is MCP only for coding? No. It connects agents to any tool that offers a server, including docs, chat, payments and monitoring. Do I need to run my own servers? Not for common tools. Friday includes a catalog of connections. Are connections limited to paid plans? No. MCP connections are on Solo and Team. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/run-multiple-coding-agents-on-one-repo # How to run multiple coding agents on one repo By [Arun Dass](/about#arun) · Published 1 October 2026 Keep agents from overwriting each other by giving each one a separate task and, where you can, separate files. Then make it easy to see who is changing what. ## What goes wrong? Two agents edit the same file. One undoes the other’s work. Or both fix the same bug in different ways. None of it shows until you try to merge. ## How do you keep them apart? 1. Split by task. Give each agent one clear job. 2. Split by files or folders where you can. 3. Use a branch for each agent when the changes are large. 4. Name each agent by its job. 5. Merge one at a time and run the tests in between. ## How do you see who is changing what? In a Friday [shared session](/shared-sessions), each agent works in its own lane, and Friday keeps track of which lane has changed which file. On one computer, only one agent can change a given folder at a time: a second agent joins in Plan, or gets its own worktree. When lanes working in separate copies change the same file, Friday warns you and says who else changed it and when. ## What if two jobs need the same file? Let one agent finish, then [hand off](/guides/ai-agent-handoff) to the next. Two agents editing one file at once is the case to avoid. ## Common questions How many agents can run at once? Up to 13 agents in one session on every plan, and your own coding tools are never metered. On Team, Friday’s own agent joins too, with up to 10 more beside it. Can agents still conflict in separate lanes? Yes, if they touch the same file. Friday gives a heads-up early. Does a person need to watch? Each agent has one driver, and anyone in the session can watch and step in. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/human-in-the-loop-coding-agents # Human in the loop for coding agents By [Arun Dass](/about#arun) · Published 1 October 2026 Human in the loop means a person can see what an agent is doing, steer it, and stop it. For coding agents, that means visible steps, a way to redirect, limits on what the agent may do, and review before anything ships. ## What controls should you look for? Control What it does See every step Shows what the agent is doing as it happens Redirect Anyone can suggest a change. The driver sends it to the agent Interrupt The driver can stop a run without restarting it Limits Sets what the agent is allowed to do Review Lets you accept a change or send it back Undo Rolls a run back ## How do you set limits in Friday? When you link a machine, you choose a mode: plan only, edit files, or edit files and run commands. Friday does not go beyond the mode you pick. The session owner can pause or cancel a run at any time, even while someone else is driving, and chooses which connections agents can use. ## When should a person approve something? - Anything that deletes or overwrites data. - Anything that touches production. - Anything that spends money or sends messages to other people. ## What does human in the loop not do? It does not make an agent correct. A person still has to look. See [how to review AI-generated pull requests](/guides/review-ai-generated-pull-requests). ## Common questions Does someone need to watch every step? No. Set limits, review before anything ships, and watch closely when the work is risky. Can I stop an agent mid-run? Yes. Whoever is driving can interrupt a run without restarting it, and the session owner can pause or cancel one at any time. Who can pause or cancel a run? The session owner always can, even when someone else is driving. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/shared-control-of-ai-agents # Shared control of AI agents: the driver model By [Arun Dass](/about#arun) · Published 1 October 2026 In a shared session, each agent has one driver at a time: the person whose instructions it follows. Everyone else can watch, suggest or ask to take over. One driver keeps an agent from getting two sets of orders. ## Why one driver? If two people prompt an agent at once, it gets mixed instructions. A single driver keeps its direction clear, and the rest of the team still takes part. ## What can everyone else do? - Watch every step. - Suggest a redirect. The driver sees it and can send it to the agent. - Request control. Only the driver can instruct, interrupt, pause or cancel the agent. The session owner can pause or cancel at any time, even when someone else is driving. ## How does control change hands? Control can pass from one person to another. A session is set to ask, so the driver approves a request, or to take over, so anyone allowed to drive can grab control. The agent keeps its context, and the log records who was driving. If the driver disconnects, control passes to the person who has been waiting longest, or is released. Each agent has its own driver, so one person can steer one agent while a teammate steers another. ## Where does this fit? This is how a [shared session](/shared-sessions) works, and a core part of [multiplayer AI](/guides/multiplayer-ai). It also maps onto [pair programming](/guides/ai-pair-programming-with-agents), with the driver and navigator roles. ## Common questions Can different people drive different agents? Yes. Each agent has its own driver. Who can pause an agent? The driver, and the session owner at any time, even when someone else is driving. Does the agent lose context when control changes? No. The agent keeps its context, and the session log records who was driving. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/share-context-between-coding-agents # Why coding agents need shared context By [Arun Dass](/about#arun) · Published 3 October 2026 Every coding agent starts knowing only what you tell it. When you move a task from Claude Code to Codex, or from one teammate to another, the context stays behind unless something carries it. Sharing context is what makes a second agent, or a second person, useful on the first one’s work. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## What is context, for a coding agent? Context is everything the agent has learned about the task: what you asked, what it read, what it tried, what failed and why, and what you decided along the way. It is the difference between “fix the login bug” and “fix the login bug, which is not in the session code, we already ruled that out”. Most of it lives in one conversation, in one tool, on one person’s machine. ## What goes wrong when context is not shared? - You switch from Claude Code to Codex and explain the task again from scratch. - A teammate picks up your work and repeats the investigation you finished an hour ago. - Two agents make different assumptions about the same code, because neither saw the other’s findings. - Nobody can review a change without reading a whole private transcript, or guessing. Each of these costs time, and each also costs tokens, because an agent that starts cold re-reads and re-tries what the last one already did. ## What does it look like when it is shared? Situation Without shared context With one shared session Switching from one coding tool to another Re-explain the task The next agent works from the same session record A teammate joins Read a transcript or ask you They see every step, and catch up from a short digest Handing work to a person or another agent A pasted summary A handoff that carries the task and its context Reviewing what happened Dig through a private chat One record of who asked what, and what each agent did ## How does Friday share context? Friday puts people and agents in one session. Everyone in it, and every agent, works from the same record: the messages, the tool calls and the file changes. You can run Claude Code, Codex, Cursor and Friday’s own agent in the same session, each in its own lane, and hand work from one to the next. The agents themselves still run on your own machine and your own accounts. ## What should you still pass on yourself? Shared context does not replace a clear instruction. Say what “done” means, name the files that matter, and write down any decision an agent could not know. A short note at the start of a lane is worth more than a long transcript. ## Common questions Is shared context the same as a bigger context window? No. A larger window lets one agent hold more. Shared context is about the same information reaching the next agent or person, instead of staying in one conversation. Do different AI models see the same context in Friday? They work from the same session. Each agent keeps its own way of reading it, so a clear instruction at the start of each lane still helps. Does sharing context send my code to Friday? Your coding agents run on your machine. The session record is what your team shares. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/continue-claude-code-session-with-your-team # How to continue a Claude Code or Codex session with your team By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 13 October 2026 You do not have to start over to bring your team in. Friday lists the Claude Code and Codex sessions already on your machine, and you can pick one and continue it, with the memory it built, in a shared session your teammates can join. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## Why continue an existing session? The hard part of a long agent session is the context in it: what the agent read, what it ruled out, what you agreed. Starting a new session to share it throws that away. Continuing the one you have keeps it. ## How do you bring a session into Friday? 1. Link your machine. Friday finds Claude Code and Codex on your computer. 2. Open the list of sessions your coding agent already has and pick the one you were in. 3. Friday shows the room what was said, then continues that conversation, so your next instruction lands in the session it already had. 4. Invite your team by link, as editors or viewers. 5. Teammates watch each step, suggest changes, or ask to take control. Work from the session lands on a new branch named after what you asked for. ## What comes with the session, and what does not? Item Claude Code and Codex The conversation and the memory the tool built Continued, so the agent still has it Where it is stored On your machine. Friday reads the files to list them and does not copy the conversation elsewhere Your agent definitions Import from Claude Code, Cursor and Codex, as private agents you can then share Cursor sessions Read in as history, so Cursor, or any other coding tool, can carry on from them Switching tools Any coding tool can carry on from another’s session. The same tool resumes its own ## What can your team do once they are in? - Watch every step as it happens, including tool calls and file changes. - Suggest a redirect for the driver to send. - Ask for control, which the driver approves or hands over, depending on the session. - Catch up from a short digest instead of reading the whole history. ## Common questions Does importing copy my conversation to Friday? No. Friday reads the session files on your machine to list them and resumes the session by its id. It does not copy the conversation anywhere. Can I import a Cursor session? Yes. Friday reads a Cursor chat in as history, the same way it reads Claude Code and Codex sessions, so any coding tool can carry on from it. Can teammates continue my session on their machine? They join the shared session and steer it. The agent keeps running on the machine it was linked on. How do we know how much room the agent has left? Claude Code and Codex limits show beside the session, for everyone in it. See [Claude Code and Codex usage limits in one place](/guides/claude-code-and-codex-usage-limits). ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/share-tools-between-coding-agents # Connect your tools once, use them with every coding agent By [Arun Dass](/about#arun) · Published 3 October 2026 Each coding agent has its own way of connecting to tools, so teams end up setting up GitHub, Slack and the rest again for every agent and every person. In Friday a workspace admin connects a tool once, and you choose which agents may use it. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## What goes wrong when every agent has its own setup? - The same tool is configured separately in Claude Code, Codex and Cursor. - Each teammate sets it up again on their own machine. - Nobody knows which agent can reach which tool, or who approved it. ## How does it work in Friday? MCP, the Model Context Protocol, is the open standard for connecting agents to tools and data. Friday has a catalog of 80+ connections, including Google Workspace, Slack, GitHub, Linear, Notion and Stripe. 1. A workspace admin adds a connection once. 2. Its tools are offered to sessions in the workspace. 3. You limit each agent to all connections, none, or a chosen few. A session can switch connections off. 4. Every tool call shows up in the session feed, so the team sees what an agent used and when. ## How should a team decide what to connect? Connect only what the work needs. An agent can use a connected tool within the permissions you approved, including tools that change data, so start with read access and add more when a task needs it. Keep the connection list short and review it when the team changes. ## Common questions Do all agents in a session use the same connections? Connections belong to the workspace, and each agent has its own allowed list, so you can give one agent GitHub and another none. Are MCP connections extra on a paid plan? No. They are included on Solo and Team. Can I see what an agent did with a tool? Yes. Tool calls appear in the shared session feed. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/claude-code-permissions-for-teams # Claude Code permissions for teams: Plan, Manual and Auto By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 13 October 2026 A team needs to decide what a coding agent may do without asking, and who answers when it does ask. In Friday every session runs in one of three modes, Plan, Manual or Auto, and approvals are visible to everyone in the session. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## What are the three modes? Mode What the agent may do Use it when Plan Read the code and propose a plan. It asks before it runs a command, and changes nothing until the driver approves the plan You want to agree the approach before any file changes Manual Make changes, asking first: each edit or command waits for approval, and the approval shows exactly what it wants to do Everyday work where you want a person in the loop Auto Edit files and run commands without stopping to ask A task you trust, in a folder you can reset The mode applies to the agent you are running, whether that is Claude Code, Codex, Cursor or Friday’s own agent. ## Who approves a command? Each agent has one driver at a time. An approval goes to the driver of the agent that asked, and the session record shows who approved what and when. Teammates watch the same approval card, so a second pair of eyes sees what is about to run before it does. A teammate can ask for control, or suggest a redirect, but they cannot make an agent run something the driver has not approved. ## How should a team set it up? 1. Start new work in Plan, and read the plan together. 2. Move to Manual to make changes. Approve them one at a time, and read what each one does. 3. Use Auto only for a bounded task in a folder you can reset, and check the receipt afterwards. 4. Keep one person as the driver of each agent. After a run, the receipt lists the commands it really ran and whether they passed, and flags a test that was weakened or a pass that never ran. ## How do these compare with Claude Code’s and Codex’s own modes? Each tool has its own settings, and the same word can mean different things. Friday’s Plan is close to Claude Code’s plan mode, which reads and proposes and does not edit your source until the plan is approved. Friday’s Manual is close to Claude Code’s default mode, where edits and commands ask first. Auto differs most. In Friday it runs without asking. Claude Code’s auto mode still has a classifier check each action, and Codex’s Auto still asks before network access or edits outside the workspace. Codex’s never approval policy is the nearest match to Friday’s Auto. ## Sources Checked against each maker’s documentation on 13 October 2026. These tools change often, so check the latest version. - [Choose a permission mode](https://code.claude.com/docs/en/permission-modes), Claude Code docs - [Agent approvals and security](https://learn.chatgpt.com/docs/agent-approvals-security), Codex docs ## Common questions Do the modes work with Claude Code, Codex and Cursor, not only Friday’s own agent? Yes. Plan, Manual and Auto all work with Claude Code, Codex and Cursor, so the team gets the same approvals whichever tool is running. Can a viewer approve a command? No. Approvals go to the driver. Viewers can watch. Is there a record of approvals? Yes. The session log records who asked, what the agent did, and who approved it. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/claude-code-and-codex-usage-limits # See your Claude Code and Codex usage limits in one place By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 13 October 2026 Claude Code and Codex each have their own usage windows, and each reports them in a different place. Friday reads both on your machine and shows them beside your session, so a team can see how much room is left before an agent stops. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## Why are usage limits hard to follow across tools? On a Claude plan, Claude Code works in a rolling five-hour window with a weekly cap. On a ChatGPT plan, Codex has its own five-hour and weekly windows, though some plans skip the five-hour one. You check one in a terminal command or settings page, and the other somewhere else. When several people and several agents share work, nobody sees the whole picture until an agent stops mid-task. ## What does Friday show? - A context ring on the session that shows how full the agent’s context is. - A usage panel with a tab for Friday, Claude Code, Codex and Cursor. - For Claude Code and Codex, the limit windows your own account reports, read on your machine. - For Friday’s own model, credits in 5-hour windows, and when the window resets. ## What should a team do with it? 1. Check the panel before starting a long run. 2. Give the long task to the agent with the most room left. 3. Hand the work to another agent in the same session when one runs low, on any plan. Your coding agents run on your own accounts, so their limits are your provider’s limits. Friday never meters them. ## Sources Checked against each maker’s documentation on 13 October 2026. These tools change often, so check the latest version. - [Use Claude Code with your Pro or Max plan](https://support.claude.com/en/articles/11145838-using-claude-code-with-your-pro-or-max-plan), Claude Help Center - [Manage costs effectively](https://code.claude.com/docs/en/costs), Claude Code docs - [Pricing](https://learn.chatgpt.com/docs/pricing), Codex docs ## Common questions Does Friday change my Claude Code or Codex limits? No. They come from your own accounts. Friday only shows them. Where does the data come from? It is read on your machine from what the tools themselves record. Nothing is estimated on a server. What happens when one agent runs out? The session keeps going. Hand the task to another agent in the session, or wait for the window to reset. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/claude-code-for-teams # Claude Code for teams: one live session for people and agents By [Arun Dass](/about#arun) · Published 1 October 2026 Claude Code is built for one person at a time. Friday Work puts your team in the same session: teammates join live, your Claude Code stays on your machine, and the session you already have carries on with its memory. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## What does a team still need around Claude Code? What the team needs Claude Code on its own Claude Code in Friday A teammate joins a running session Share links are read-only They join live, watch and steer Pick up a session someone started Export and re-explain Continue the session, with its memory Use Codex or Cursor on the same task Separate tools, separate context Side by side in one session Approve what an agent runs Per person, in their terminal Plan, Manual or Auto, with approvals the team can see Share tools like GitHub and Slack Set up per agent and per person Connect once, choose which agents may use it See what happened A private transcript One record of who asked what and what each agent did ## How do you start? 1. Sign in to Friday in your browser, or in the optional desktop app for Windows or Mac. 2. Link your machine with one command, or let the desktop app do it. Friday finds Claude Code, Codex and Cursor. 3. Pick a Claude Code session you already have, or start a new one. 4. Invite your team by link. Claude Code runs on your own account, and Friday never meters it. Choose what it may do without asking in [Claude Code permissions for teams](/guides/claude-code-permissions-for-teams), and see how much room is left before it stops in [usage limits in one place](/guides/claude-code-and-codex-usage-limits). ## When is Claude Code on its own enough? If you work alone, or only need to show someone what happened, Claude Code and its share link are simpler. Friday is for when more than one person needs to act in the same session, or when you use more than one coding tool. ## Common questions Does Friday replace Claude Code? No. It runs your own Claude Code and gives your team a shared session around it. Do I need a Claude Code subscription? Yes. It runs on your own account, and Friday does not meter it. What does it cost? Solo is free for one person. Team Basic is $15 a seat a month and Team Pro is $30, or $12 and $25 paid yearly. New teams get Team Basic free for 7 days. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/aq # Friday Work vs AQ for team coding agents By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 14 October 2026 Both put a team and coding agents in one shared place. AQ describes itself as a multiplayer coding harness: agents run on a cloud machine, AQ’s or your own, and the team works in the browser. Friday Work runs your own Claude Code, Codex and Cursor on your own machines and shares the session around them. ## How do they differ? Friday Work AQ (from its own site) Where agents run On each person’s own machine and accounts On a cloud machine: an AQ-managed one, or on Team, one in your own cloud What is shared The live session, record, approvals and tools A live terminal, code editor and app preview; teammates can type in or take over Existing sessions Continue Claude Code and Codex sessions you already have Not described on its site Tools Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent Claude Code, Codex, Cursor and 8 more, plus its own agent Platforms In the browser, plus a desktop app for Windows and Mac In the browser, on any computer or phone Last checked: 14 October 2026 Based on AQ’s own site and docs. Check aq.dev for its current features and prices. ## When is AQ the better pick? If you want agents to run in a managed cloud environment, with live terminals, an editor and previews of the running app in one browser tab, and you are happy for the code to run there, AQ’s model fits. Friday keeps the work on your machines. ## When is Friday the better pick? If your code should stay on your own machines, if people already work in their own Claude Code or Codex sessions, or if you want a team to approve what an agent runs, Friday fits. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## Common questions Is Friday Work a cloud IDE? No. Friday works in your browser, with an optional desktop app, but your coding agents run on your own machine, and the session is shared around them. Can I use both? They solve similar problems differently, so most teams choose one. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/nimbalyst # Friday Work vs Nimbalyst for teams using Claude Code and Codex By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 14 October 2026 Both keep agents local and add a shared space for the team. Nimbalyst describes a multiplayer visual workspace for docs, mockups, diagrams and trackers. Friday Work focuses on the coding session itself: people and agents steering the same task live. ## How do they differ? Friday Work Nimbalyst (from its own site) Center of the product A live shared coding session A shared visual workspace around the work Where agents run On your own machine and accounts On each teammate’s own machine What teammates edit together The session: instructions, approvals, control Docs, mockups, diagrams and trackers Tools Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent Claude Code and Codex; OpenCode and Copilot in alpha Existing sessions Continue Claude Code and Codex sessions you already have Imports Claude Code sessions Approvals Plan, Manual or Auto; risky commands wait for a yes, in front of the team Each AI edit shown as a diff to approve Platforms In the browser, plus a desktop app for Windows and Mac Mac, Windows and Linux, plus an iOS app Last checked: 14 October 2026 Based on Nimbalyst’s own site and docs. Check nimbalyst.com for its current features and prices. ## When is Nimbalyst the better pick? If your team’s shared work is mostly written and visual, such as specs, mockups and diagrams edited together alongside coding agents, its workspace is built for that. ## When is Friday the better pick? If the shared work is the coding session itself, with several people steering agents, approving commands and taking control in turn, Friday is built for that. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## Common questions Do Friday and Nimbalyst do the same thing? They overlap on shared work around local agents. Friday centers on the live coding session, and Nimbalyst on a visual workspace. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/alternatives/conductor-windows # A Conductor alternative for Windows, and for teams By [Arun Dass](/about#arun) · Published 3 October 2026 · Updated 14 October 2026 Conductor runs parallel coding agents from a Mac app, on your Mac or in its cloud. If you are on Windows, or you want your team in one live session while the agents keep running on your own machines, Friday Work is built for that. It works in your browser, with an optional desktop app for Windows and Mac. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## How does Friday compare with Conductor? Friday Work Conductor (from its own site) Platforms In the browser, plus a desktop app for Windows and Mac Mac, with cloud workspaces run from the Mac app Agents Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent Claude Code, Codex, Cursor and OpenCode Several agents on one task Side by side in one shared session Parallel agents, each in its own git worktree or cloud sandbox Teammates Join a live session by link, steer, and take over; watching is free On paid plans: shared cloud workspaces with presence and live updates Approvals Plan, Manual or Auto; risky commands wait for a yes, in front of the team Review each diff before a PR; approve a plan in Plan mode Last checked: 14 October 2026 Based on Conductor’s own site and docs. Check conductor.build for its current features and prices. ## When is Conductor the better pick? If you work on a Mac and want many agents running in separate worktrees or cloud sandboxes, with a diff-first review screen, Conductor is simpler. ## When is Friday the better pick? If you are on Windows, if your team needs to steer and take over agents that run on your own machines, or if you use coding agents beyond Claude Code, Codex, Cursor and OpenCode, Friday fits. ## Common questions Does Friday work on Windows? Yes. Friday works in your browser, and the optional desktop app runs on Windows and Mac. Can Friday run several agents on one repo? Yes. Agents work in their own lanes. On one computer only one agent changes a folder at a time, so a second joins in Plan or gets its own worktree; lanes in separate copies get a warning when they change the same file. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/conductor # Friday Work vs Conductor By [Arun Dass](/about#arun) · Published 14 October 2026 Both run several coding agents at once. Conductor runs them on a Mac or in its cloud, and its team features live in that cloud. Friday runs them on your own computer, in one live session your team can watch, steer and take over. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## How do Friday and Conductor compare? Friday Work Conductor (from its own site) Platforms In the browser, plus a desktop app for Windows and Mac Mac app, plus a mobile app on Pro Coding agents Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent Claude Code, Codex, Cursor and OpenCode Where agents run On your own computer, under your own accounts On your Mac, or in Conductor’s cloud on paid plans Working as a team One live session: watch, steer and take over Shared cloud workspaces and chats, on Pro and Teams Who can watch Anyone with the link, free, on any plan Members of your Conductor organization Approvals Plan, Manual or Auto; risky commands wait for a yes Plan mode, and diff review before a pull request Existing sessions Continue a Claude Code, Codex or Cursor session Not described on its site Price Free for one; $15 or $30 a seat a month for teams Free; Pro $50 a month; Teams $60 a user a month Enterprise Not built for enterprise: no SSO or SOC 2 SOC 2 Type II; SAML SSO and SCIM on Enterprise Last checked: 14 October 2026 Based on Conductor’s own site, pricing page and docs. Check conductor.build for its current features and prices. ## How does working together differ? In Conductor, team work happens in cloud workspaces. The agent runs in a sandbox Conductor hosts, the chat updates live for everyone in your organization, and you can reassign a workspace to a teammate. In Friday, the agents stay on the computer of whoever is driving. The team joins that live session, one person drives each agent, and it hands on to a teammate or another agent with its full history. ## Where do your code and your session go? - Conductor on your Mac: messages go straight to your model provider, and session data stays on your device. - Conductor Cloud: the agent works in a sandbox in Vercel’s us-east-1 region, and chat messages are stored on Conductor’s servers. - Friday: your repository stays on your computer and is not copied. The session itself, its messages, steps and commands, goes through Friday’s servers so your team can watch it. [What Friday stores](/security#collects). ## What does a team of five pay? Plan A month, five people What the team gets Friday Team Basic $75 One live session for the team, any of 14 agents Friday Team Pro $150 The same, plus frontier models for Friday’s own agent Conductor Pro $250 Cloud workspaces, live collaboration for up to 5 Pro users Conductor Teams $300 Collaboration at any size, admin portal, central billing Monthly list prices. Both use the Claude, ChatGPT or Cursor plans you already pay for, on top. People who only watch a Friday session need no seat. ## When is Conductor the better pick? If your team is on Macs and wants agents that keep running in the cloud after you close the app, with a diff-first review before each pull request, Conductor fits. It also has an API and a mobile app, and it offers SOC 2 and SSO for companies that need them. ## When is Friday the better pick? If anyone on the team uses Windows, if your code should stay on your own computers, or if you want people to watch for free and step in live, Friday fits. It also runs more coding agents, for less a seat. ## Common questions Does Conductor run on Windows? Its site offers a Mac app, plus a mobile app on Pro. Friday works in your browser, or in its optional desktop app for Windows and Mac. Is Conductor multiplayer? Yes, on its Pro and Teams plans: cloud workspaces and chats are shared within your organization. Friday shares a live session on your own computer, and anyone with the link can watch. Does Friday run agents in the cloud? No. Your coding agents run on your own computer, under your own accounts. Friday shares the session around them. Which is cheaper for a team? At list prices, five people cost $75 or $150 a month on Friday, and $250 or $300 on Conductor. Both are on top of your own agent plans. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/claude-code-vs-codex-vs-cursor # Claude Code vs Codex vs Cursor: which coding agent should you use in 2026? By [Arun Dass](/about#arun) · Published 12 October 2026 Claude Code, Codex and Cursor all start at $20 a month and can all edit code, run commands and work across a repository. Claude Code is Anthropic’s agent, built around the terminal and Claude models. Codex is OpenAI’s, included with ChatGPT plans. Cursor is a full code editor with an agent built in and models from several labs. ## Claude Code vs Codex vs Cursor at a glance Claude Code Codex Cursor Made by Anthropic OpenAI Anysphere What it is A coding agent you run in the terminal, your IDE, a desktop app or the web A coding agent in a CLI, IDE extension, desktop app and cloud tasks A code editor with an agent built in, plus a CLI and cloud agents Models Anthropic’s Claude models OpenAI’s GPT models Models from several labs, plus Cursor’s own Cheapest plan with the agent Claude Pro, $20 a month ($17 billed yearly) ChatGPT Plus, $20 a month for the CLI, IDE and cloud; Free and Go get the desktop app Pro, $20 a month; the free Hobby plan has limited agent requests Team plan Team, $25 a seat a month ($20 yearly) Business, $25 a user a month ($20 yearly) Teams, $40 a user a month How limits work Usage per five-hour session; more on Max and Team Premium seats Limits set by plan and shared with ChatGPT Included usage, then billed on demand Project instructions file CLAUDE.md, and AGENTS.md when there is no CLAUDE.md AGENTS.md Rules in .cursor/rules, and AGENTS.md Last checked: 12 October 2026 Prices are from each maker’s own pricing page on 12 October 2026, in US dollars. All three change their plans often, so check the sources at the end of this page before you buy. ## Claude Code vs Codex Both are agents first: you describe a task and they plan, edit files and run commands until it is done. The biggest difference is the model behind each one, Claude for Claude Code and GPT for Codex, and teams often find one suits their codebase better after a week of real tasks. - Choose Claude Code if you live in the terminal, want fine control through skills, hooks and subagents, or your team is already on a Claude plan. - Choose Codex if you already pay for ChatGPT, want to hand work to cloud tasks and check it from your phone, or prefer GPT models for your stack. - Both read an AGENTS.md file, so one set of project instructions can serve both. ## Claude Code vs Cursor Claude Code is an agent you add to the tools you already use. Cursor replaces your editor: you get tab completion, inline edits and an agent panel in one app, with a choice of models that includes Claude. - Choose Claude Code if you want an agent that works the same in any editor or none, and the deepest support for Claude models. - Choose Cursor if you want completions and an agent in one editor, or want to switch between models from different labs without leaving it. - You can run both: Claude Code has an extension for VS Code-based editors, which includes Cursor. ## Codex vs Cursor Codex comes with ChatGPT, so many people already have it. Cursor costs the same for one person but more for a team, $40 a user against $25, and gives you a whole editor for it. - Choose Codex if your team already has ChatGPT Business and wants an agent without changing editors. - Choose Cursor if your team wants one editor everyone shares, with team rules, code review and usage analytics. ## What about Gemini CLI and OpenCode? Gemini CLI is Google’s open-source terminal agent, built around Gemini models. OpenCode is an open-source terminal agent that works with many model providers, including models you run yourself. Both are worth a look if you want an open-source agent or already have the models they use. Both also read AGENTS.md, according to the AGENTS.md project. ## Which one should you pick? 1. Start with the plan you already pay for. A Claude plan includes Claude Code and a ChatGPT plan includes Codex, so the cheapest agent is often the one you have. 2. Try each on the same real task in your own repository for a few days. Benchmarks rarely match your codebase. 3. Pick by how you work: terminal and fine control (Claude Code), cloud tasks and ChatGPT (Codex), or one editor with completions (Cursor). 4. Keep your project instructions in AGENTS.md, so switching later costs nothing. ## Can you use Claude Code, Codex and Cursor together? Yes, and many developers do: one agent writes, another reviews, and the third takes over when the first hits its usage limit. The friction is keeping them in step, since each has its own window, history and limits. Friday Work runs Claude Code, Codex, Cursor and 11 more coding agents side by side in one session, on your own plans. Smart routing suggests the agent and model most likely to get each task right for the least, every tool’s usage limits sit in one place, and a project brain briefs whichever agent comes next. It is free for one person. See [how to use them together](/guides/use-claude-code-codex-and-cursor-together). ## How we compared them We read each maker’s pricing page and documentation on 12 October 2026 and compared what each says about itself. We build Friday Work, which runs all three, so we have no reason to favor one. Where a fact depends on a plan or version, the table says so. - [Claude pricing](https://claude.com/pricing), Anthropic - [Codex pricing](https://learn.chatgpt.com/docs/pricing), OpenAI - [Cursor pricing](https://cursor.com/pricing), Anysphere - [AGENTS.md](https://agents.md/), the Agentic AI Foundation ## Common questions Is Claude Code better than Codex? Neither wins everywhere. Claude Code gives more control through skills, hooks and subagents and runs Claude models; Codex comes with ChatGPT and adds cloud tasks. Try both on the same task in your own repository. Is Codex free? Partly. ChatGPT Free and Go include Codex in the desktop app on a smaller model, subject to rollout. The CLI, IDE extension and cloud tasks start with ChatGPT Plus at $20 a month. Is Claude Code free? No. Claude Code starts with Claude Pro at $20 a month, or $17 a month billed yearly. The free Claude plan does not include it. Does Cursor use Claude? Cursor lets you choose models from several labs, and Claude models are among them. It also has models of its own. Can I use Claude Code inside Cursor? Yes. Claude Code has an extension for VS Code-based editors, and Cursor is one, so you can run both in the same window. Which is cheapest for a team? Claude Team and ChatGPT Business are both $25 a seat a month, or $20 billed yearly. Cursor Teams is $40 a user a month. Can one AGENTS.md file work for all three? Yes. Codex and Cursor read AGENTS.md, and Claude Code reads it when the project has no CLAUDE.md, or you can import it from your CLAUDE.md. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/agents-md-vs-claude-md # AGENTS.md vs CLAUDE.md: which file do your coding agents read? By [Arun Dass](/about#arun) · Published 12 October 2026 · Updated 13 October 2026 AGENTS.md is an open format that most coding agents read for project instructions, including Codex, Cursor, GitHub Copilot and Gemini CLI. CLAUDE.md is Claude Code’s own file. Claude Code now reads AGENTS.md too, when a project has no CLAUDE.md, so one file can brief every agent your team uses. ## What is AGENTS.md? AGENTS.md is a Markdown file at the root of a repository that tells coding agents how to work in it: how to build and test, the code style, and anything a new contributor would need. Its own site calls it “a README for agents”. It has no required fields. The format grew out of work between OpenAI Codex, Amp, Google’s Jules, Cursor and Factory, and is now stewarded by the Agentic AI Foundation under the Linux Foundation. In a monorepo you can put an AGENTS.md in each package, and agents read the nearest one, so the closest file to the code being edited wins. ## What is CLAUDE.md? CLAUDE.md is Claude Code’s memory file: plain Markdown that Claude Code reads at the start of every session. It can live in several places, and Claude Code loads all of them, broadest first. File Where Who it is for Project ./CLAUDE.md or ./.claude/CLAUDE.md The whole team, through the repository Local ./CLAUDE.local.md (add it to .gitignore) Only you, in this project User ~/.claude/CLAUDE.md Only you, in every project Managed policy A system folder set by IT Everyone in the organization Claude Code also reads CLAUDE.md files in folders above where you start it, and loads the ones in subfolders when it works there. A CLAUDE.md can pull in other files with @path, for example @docs/testing.md. ## Which file does each coding agent read? Agent Reads Notes Claude Code CLAUDE.md, or AGENTS.md when there is no CLAUDE.md Reading AGENTS.md directly needs version 2.1.277 or later Codex AGENTS.md Also ~/.codex/AGENTS.md for you, and AGENTS.override.md to override a folder Cursor .cursor/rules/\*.mdc and AGENTS.md The Cursor CLI also reads a CLAUDE.md at the project root GitHub Copilot, Gemini CLI, Junie, Amp, OpenCode, Kilo Code AGENTS.md Listed as supporting it on agents.md Checked against each maker’s documentation on 12 October 2026. ## How do you use one file for every agent? 1. Write your project instructions in AGENTS.md at the root of the repository and commit it. 2. If you do not need Claude-only instructions, stop there: Claude Code reads AGENTS.md when there is no CLAUDE.md. 3. If you do, create a CLAUDE.md that starts with the line @AGENTS.md, then add the Claude-only parts below it. Claude Code reads both; other agents read AGENTS.md. 4. In a monorepo, add an AGENTS.md to each package that needs its own commands or rules. 5. Keep personal preferences out of the shared file: CLAUDE.local.md and ~/.codex/AGENTS.md are for you alone. A CLAUDE.md that shares AGENTS.md with every other tool: ``` @AGENTS.md ## Claude Code only - Use the test-writer subagent for new tests. - Run /review before opening a pull request. ``` ## What should go in AGENTS.md or CLAUDE.md? The AGENTS.md project suggests a project overview, build and test commands, code style, testing instructions and security considerations. Write what an agent would get wrong without being told, and leave out what it can read from the code. ``` # AGENTS.md ## Setup - Install: pnpm install - Dev server: pnpm dev ## Tests - Run pnpm test before every commit; all tests must pass. - Tests sit next to the file they cover, as name.test.ts. ## Style - TypeScript strict mode. No default exports. - Use the logger in src/log.ts, never console.log. ## Don’t - Edit files in src/generated/; run pnpm codegen instead. ``` - Keep it short. Anthropic suggests under 200 lines for a CLAUDE.md, because every line is loaded into every session. - Be specific: “run pnpm test” beats “test your changes”. - Remove rules that contradict each other; an agent may follow either one. - Run /init in Claude Code to draft a CLAUDE.md from your codebase, then edit it. ## How does Friday Work use AGENTS.md? Friday Work runs Claude Code, Codex, Cursor and 11 more coding agents side by side in one session. Each reads the project’s instruction files as it normally would. Friday adds a project brain: what agents learn on a repository, from every tool, briefs the next agent that works on it. ## Sources - [AGENTS.md](https://agents.md/), the Agentic AI Foundation - [How Claude remembers your project](https://code.claude.com/docs/en/memory), Claude Code docs - [Custom instructions with AGENTS.md](https://learn.chatgpt.com/docs/agent-configuration/agents-md), Codex docs - [Rules](https://cursor.com/docs/rules), Cursor docs ## Common questions Does Claude Code read AGENTS.md? Yes, from version 2.1.277. It reads AGENTS.md when your project has no CLAUDE.md or CLAUDE.local.md. If you have a CLAUDE.md, add @AGENTS.md to it to load both. Should I rename CLAUDE.md to AGENTS.md? If your team uses more than one coding agent, yes: move the shared instructions to AGENTS.md and keep a short CLAUDE.md that imports it, for anything only Claude Code needs. Can I have both AGENTS.md and CLAUDE.md? Yes. Claude Code reads CLAUDE.md and ignores AGENTS.md unless CLAUDE.md imports it with @AGENTS.md. Other agents read AGENTS.md. Where does CLAUDE.md go? At the root of the repository for the team, in ~/.claude/CLAUDE.md for your own preferences, and in CLAUDE.local.md for personal notes on one project. What is CLAUDE.local.md? Your private CLAUDE.md for one project. It loads after CLAUDE.md and should be in .gitignore so it is not shared. What are Cursor rules? Files in .cursor/rules with the .mdc extension, each with frontmatter that says when it applies. Cursor reads AGENTS.md too, as a simpler option. How long should AGENTS.md be? As short as it can be while covering what agents get wrong. Anthropic suggests under 200 lines for CLAUDE.md, and the same reasoning holds for AGENTS.md. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/claude-code-skills-for-teams # Claude Code skills: what they are, examples, and how to share them with your team By [Arun Dass](/about#arun) · Published 12 October 2026 A Claude Code skill is a folder with a SKILL.md file that teaches Claude a task: a deploy checklist, a review routine, a house style. Claude reads only each skill’s short description until the skill is needed, so you can keep many. Commit skills to your repository and everyone on the team gets them. ## What is a Claude Code skill? A skill is a directory with a SKILL.md file: a short header that says what the skill does and when to use it, then instructions in Markdown. It can hold more files, such as reference notes or scripts, that Claude opens only when it needs them. Claude uses a skill in two ways. It picks one on its own when your request matches the skill’s description, or you call it by name, for example /release-notes. Until then, only the description is in Claude’s context, which is why a skill costs almost nothing until it runs. ## Where do Claude Code skills live? Kind Path Who gets it Project .claude/skills//SKILL.md Everyone working in the repository Personal ~/.claude/skills//SKILL.md You, in every project on your machine Plugin /skills//SKILL.md Anyone who installs the plugin, as /plugin-name:skill-name Enterprise Deployed through managed settings Everyone in the organization If two skills share a name, enterprise wins over personal, and personal over project. ## How do you write a skill? 1. Make a folder in .claude/skills named for the task, such as release-notes. 2. Add a SKILL.md with a name and a description. The description decides when Claude uses it, so say what it does and when. 3. Write the steps below the header, as you would explain them to a new teammate. 4. Move long reference material into separate files and point to them from SKILL.md, which should stay under 500 lines. 5. Try it by typing /release-notes, then by asking for the task in plain words to check Claude picks it on its own. A complete skill: ``` --- name: release-notes description: Writes release notes from the commits since the last tag. Use when someone asks for release notes or a changelog entry. --- ## Commits since the last release !`git log $(git describe --tags --abbrev=0)..HEAD --oneline` ## Instructions 1. Group the commits into New, Improved and Fixed. 2. Write each line for a customer, not a developer. 3. Leave out internal refactors and dependency bumps. ``` The line starting with ! runs a command and puts its output into the skill before Claude reads it. ## Which settings matter in SKILL.md? Field What it does description What the skill does and when to use it. Claude reads this to decide. name The command name, as in /name. Defaults to the folder name. disable-model-invocation: true Only a person can run it. Use for deploys and anything risky. user-invocable: false Hidden from the / menu; Claude uses it as background knowledge. allowed-tools Tools the skill may use without asking each time. context: fork Runs the skill in a separate subagent. paths File patterns that limit when Claude loads it on its own. ## Claude Code skill examples for teams - Deploy: the exact steps and checks for shipping, with disable-model-invocation so only a person can start it. - Review against house rules: what your team checks in a pull request, so every review asks the same questions. - Release notes: commits since the last tag, grouped and written for customers, as in the example above. - Database migration: how your team names, writes and tests migrations, and what never to do in production. - Incident triage: where the logs and dashboards are, and the first five things to check. - Test writer: your test framework, file layout and the cases your team expects for every change. ## How do you share skills with your team? - Commit .claude/skills to the repository. Everyone who works in it gets the same skills, and changes go through review like code. - Package skills in a plugin to share them across many repositories, and publish it in a plugin marketplace your team installs from. - For a whole organization, deploy skills through managed settings so they apply everywhere. Personal skills in ~/.claude/skills stay on your machine, so move anything the team relies on into the repository. ## Skills vs CLAUDE.md vs subagents vs hooks Use it for When it loads CLAUDE.md Rules that apply to every task in the project Every session, in full Skill A task Claude does sometimes, with its own steps The description always; the rest when used Subagent Work that needs its own context window, such as a long review When Claude hands work to it Hook Something that must always happen, such as blocking a command On the event it is set for, every time Instructions in CLAUDE.md and skills are guidance Claude follows; a hook is enforced. If an action must never happen, use a hook. ## Do skills work in other coding agents? The SKILL.md format follows Agent Skills, an open standard, so a skill can work in other agents that adopt it. Claude Code adds options of its own, such as disable-model-invocation and context: fork, that other tools may ignore. In Friday Work, Claude Code runs as your own Claude Code, so your project and personal skills work as they do in the terminal, and its commands and skills appear in the message box. Friday’s own agents can have skills too, added in the agent builder. ## Sources - [Skills](https://code.claude.com/docs/en/skills), Claude Code docs, checked 12 October 2026 - [Agent Skills](https://agentskills.io), the open standard ## Common questions Where are Claude Code skills stored? In .claude/skills//SKILL.md for a project, ~/.claude/skills//SKILL.md for you alone, or inside a plugin. How do I run a Claude Code skill? Type / and its name, such as /release-notes, or ask for the task in plain words and Claude uses the skill if the description matches. Do skills use up context? Barely. Only each skill’s name and description stay in context; the rest loads when the skill runs. How do I stop Claude running a skill on its own? Add disable-model-invocation: true to the skill’s header. Then only a person can start it. What is the difference between a skill and CLAUDE.md? CLAUDE.md holds rules for every task and is always loaded. A skill holds the steps for one task and loads when that task comes up. Where can I find Claude Code skills? Your team’s own repository is the best source, since skills encode how you work. You can also install skills that others publish in plugins. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/claude-code-and-codex-talk-to-each-other # Claude Code and Codex talking to each other: how agents share context in Friday By [Arun Dass](/about#arun) · Published 21 October 2026 Claude Code and Codex do not message each other on their own: each keeps what it learns in its own terminal. In a Friday session every coding agent gets a server named friday, with tools to message another agent, ask it a question and wait for the answer, post updates, share notes and hand a task on. Every exchange lands in the session feed, where you can read it. ## Why can’t coding agents talk to each other by default? Each coding agent runs as its own process with its own history. While it works it can call out to tools, but nothing pushes new information into a turn that is under way, so it hears about the world again only at its next turn. A teammate, or you, ends up carrying findings between them by hand. Some tools have their own version of the idea. Claude Code’s subagents report back to the parent that started them, and its agent teams message each other, but only between Claude Code instances. Nothing in either reaches a Codex or Cursor session next to it. ## How does it work in Friday? Every coding tool Friday starts gets an MCP server named friday. MCP is the standard these tools already use to call tools, so no change to Claude Code or Codex is needed. The server’s tools do nothing on your machine. Each call goes into the session’s log, which knows which agent is calling. That puts the conversation where your team already watches the work. A message arrives as an instruction in the receiving agent’s lane: it queues behind what the agent is doing, wakes an idle agent, and is kept and replayable like everything else in the session. Tool What it does list\_agents Shows who else is in the session, whose agent it is, and whether it is working or idle message\_agent Tells another agent something, without waiting ask\_agent Asks a question and waits for the answer reply\_to\_agent Answers a question right away, mid-turn post\_update Tells every other agent something without waking any of them check\_messages Reads what is waiting: messages, questions, updates, and files another agent is also changing read\_notes, write\_note Reads and writes the session’s shared notes wait\_for\_agent Waits until another agent finishes, and returns what it said hand\_off Moves the work to another agent with the whole record ## What does a question between two agents look like? Say Claude Code is changing an API and Codex is writing its tests. Codex asks Claude Code which error shape it chose. Claude Code answers, either right away with reply\_to\_agent or at the end of the turn that handled the question. By default the asker waits up to four minutes, and the limit is 15. If the answer takes longer, the asker is told, and the answer reaches it later as a message. An agent that is busy does not stop mid-sentence for a message. Every answer from the friday server ends with how many messages and updates are waiting, and the agents are told to call check\_messages between steps of a long task. ## How do agents share notes and hand off work? Shared notes hold what every agent in the session should know, such as a decision about naming. Updates tell the others that something they build on has changed, with no interruption. Friday also tells each agent to say which files it will change before it writes, and check\_messages warns when another agent is changing a file you changed. A hand-off moves a task to another agent with its full record. When a coding tool stops because it ran out of usage, Friday hands the task to another agent that can take it. If none can, it resumes when the limit resets or asks a person. It picks an agent that can carry on now, never the same tool on the same computer, and it tries another coding tool before Friday’s own agent. You can also make any agent the lead and give it one goal. It splits the work into a task board, each task naming the files it owns and what it waits on, and the free agents pick tasks up. Another agent reviews a finished task before it counts. See [how to hand off an agent session](/guides/ai-agent-handoff), and our [hand-off benchmark](/benchmarks/handoff) for what a hand-off saved in tokens. ## What keeps agents from looping or running away? - A chain of messages from one agent to the next ends after a few hops, and two agents cannot pass the same thing back and forth without limit. - One run can send only so many messages before it is told to finish its own work or ask a person. - An agent cannot ask one that is already waiting on it, which would hang both. - An agent acts for the person behind it. A message cannot do more than that person could, and an agent you steer is yours. - Stopping an agent ends its task, not the agent. It stays on standby, ready for the next message. ## What does it not do yet? - A busy agent cannot be interrupted mid-turn. It reads its messages when it checks, or at its next turn. - Cursor uses the friday server only in Auto mode. - In Manual and Plan, some agents ask a person before their first friday tool call, as they do for any MCP tool. - Desktop apps older than this feature do not offer the tools. Your agents run on your own computer and your own plans. Friday does not meter them, and Solo is free. ## Common questions Can Claude Code and Codex really message each other? Yes, inside a Friday session. Each gets the friday MCP server and can message, ask and reply to the other. Outside a session they have no channel to each other. Do the agents decide when to talk? Mostly. Friday tells them to check messages between steps, post an update when they change something others build on, and say which files they will change. You can read every exchange in the feed. Can an agent act without the person who started it? No. An agent that is woken by another agent still acts for the person behind the first one, and cannot do more than that person could. Do I pay for the agents talking? Your own Claude Code and Codex run on your own plans, and Friday never meters them. Solo is free for one person. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/run-coding-agents-in-parallel # How to run coding agents in parallel By [Arun Dass](/about#arun) · Published 21 October 2026 To run coding agents in parallel, give each one its own task and its own working copy of the code, usually a git worktree, then watch for the case that isolation does not solve: two agents deciding the same thing differently. Worktrees stop agents overwriting each other’s files. They do not stop duplicate or conflicting work. ## What does running agents in parallel mean? Several coding agents work on one repository at the same time, each on its own job: one on the endpoint, one on its tests, one on a bug. The gain is that three jobs take about as long as the longest one. The cost is that you now have three things to watch and merge. ## What are the ways to do it? Approach What you get Where it falls short Several terminals, one folder Zero setup Agents edit the same files, and you find out at merge time A git worktree per agent Each agent has its own checkout and branch, so file edits cannot collide Setup per worktree, such as installing dependencies again; conflicts move to the merge An app that manages worktrees Worktrees created and cleaned up for you Usually built for one person; check what it shows about overlap One shared session Every agent and person in one place, with each lane’s changes visible, and a worktree per agent if you choose On one computer only one agent changes a folder at a time: a second joins in Plan or gets its own worktree. Lanes in separate copies get a warning when they change the same file ## What do worktrees not fix? - Two agents fixing the same bug in different ways, each in its own clean copy. - Two agents changing a shared interface, such as a function’s arguments, in different files. Nothing conflicts until the merge, or until the tests fail. - The review. Three branches still need three reviews, and the riskiest change deserves the most attention. - Usage limits. Three agents on one Claude or ChatGPT plan reach its limit sooner than one. The rules in [how to run multiple coding agents on one repo](/guides/run-multiple-coding-agents-on-one-repo) still apply: split by task, split by files where you can, and merge one at a time. ## How does Friday show who is touching which files? In a Friday session each agent works in its own lane, and Friday keeps track of which lane has changed which file. On one computer, only one agent can change a given folder at a time: a second agent joins in Plan, or gets its own worktree. When lanes working in separate copies change the same file, Friday warns you and says who else changed it and when. The agents see it too. Friday tells each one to say which files it will change before it writes, and an agent that calls check\_messages learns which of its files another agent is changing as well. Agents can also ask each other directly, as in the guide on how Claude Code and Codex talk to each other. You can also give an agent its own worktree. Friday makes the copy on your machine, and the agent can edit without touching the files another agent holds. Because a folder takes one writer at a time on one computer, put each agent that edits in its own worktree and use the session to see all of them at once. ## How do you set it up? 1. Write each task so it names the files or folders it owns. 2. Add each agent to the session and name its lane after its job, such as tests or endpoints. 3. Start the agents, and keep one person as the driver of each. 4. Watch the heads-up Friday gives when two lanes touch the same file. When it appears, ask one agent to finish first, then hand the work on. 5. Review and merge one lane at a time, and run the tests between merges. Or make one agent the lead and give it the goal. It splits the work into a task board and hands each task to the agent best placed for it. ## When is a worktree-only tool enough? If you work alone, a tool that creates a worktree per task and nothing else may be all you need. Orca, Synara and Orchestrator are built for that today, and our comparisons with [Orca](/compare/orca) and [Conductor](/compare/conductor) say where each fits. Friday is for when other people need to see and steer the agents, or when you want the agents to share what they learn. ## Common questions How many coding agents can I run at once? Up to 13 agents in one session on every plan, and your own coding tools are never metered. In practice your Claude or ChatGPT usage limit is the ceiling. Do I need git worktrees? Not for small changes in separate folders. For large changes in the same files, a worktree per agent keeps their edits apart, and Friday can make one for an agent. Can two agents in separate lanes still conflict? Yes, if they work in separate copies and touch the same file, or make clashing decisions. On one computer Friday lets only one agent change a folder at a time, so a second agent joins in Plan or gets its own worktree; across copies it warns you early. Does Friday run agents in the cloud? No. Your coding agents run on your own computer, under your own accounts. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/claude-code-collaboration # Claude Code collaboration: four ways to work with a teammate By [Arun Dass](/about#arun) · Published 21 October 2026 · Updated 22 October 2026 Claude Code is built for one person at a keyboard, so teams collaborate around it. There are four common ways: share a screen, paste logs and transcripts, pass the work through git, or share the session itself. Each is right for something, and each loses something. ## What are the ways to collaborate on Claude Code work? Method Teammate sees Teammate can act Record afterward Screen share or call Your screen, live Only by telling you what to type None, unless someone records it Pasted logs or a transcript What you chose to paste Only by starting their own session The paste, without the agent’s context Git branch and pull request The code changes Review and commit on top The history of commits Shared session in Friday Every step, live Suggest redirects, or take control as an editor The session record, replayable ## When is screen sharing enough? For a quick look together while both people are online, a call with a shared screen is hard to beat. It costs nothing and needs no setup. It stops working when the teammate is in another time zone, when a third person wants to follow, or when somebody asks next week why the agent made a change. Nobody but you can steer, and nothing is kept. ## What goes wrong with pasted logs? A pasted transcript shows what happened but not what the agent knows. The teammate who starts a fresh session re-reads the same files and tries what you already ruled out. You choose what to paste, so the dead ends and the reasons are usually the parts that get left out. Claude Code’s own share links are read-only too, as covered in [how to share a Claude Code session with your team](/guides/share-claude-code-session-with-your-team). They suit a record, such as a bug report. They do not let anyone act. ## What does git give you, and what does it leave out? Git is the right channel for the result. A branch and a pull request carry the code, the review and the history, and every team already has them. Git does not carry the conversation behind the change. The prompt, the files the agent read and the approaches it dropped stay on one person’s machine, so a reviewer sees a diff and has to guess the intent. ## What does a shared Friday session add? Friday runs your own Claude Code on your machine and opens the session to your team. You send teammates a link, as viewers or editors. A viewer watches each step in a browser with no account and no install. Editors sign in, and can suggest a redirect for the driver to send, or ask for control. The record stays, and anyone can replay it. It costs something too. The agent keeps running on the machine it was linked on, so that machine has to stay on. Friday stores the session record so your team can see it, and it is not a pure peer-to-peer channel. The security page lists what is stored and what is not finished yet. Solo is free for one person, and teammates who steer with you need Team, from $15 a seat a month, with a 7-day Team Basic trial that starts when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent. ## Which should you use? 1. To show a result: a pull request, with the prompt and findings in its description. 2. To get a quick second opinion while you are both online: a screen share. 3. To keep a record of one run: a read-only share link or an exported transcript. 4. To work on the same task, hand it over, or let a teammate catch up later: a shared session. If you already have a session going, you do not need to start over. See [how to continue it with your team](/guides/continue-claude-code-session-with-your-team). ## Common questions Can two people use one Claude Code session at the same time? Not in Claude Code on its own. In a Friday session, teammates watch live, suggest redirects, and take control in turn. One person drives each agent at a time. Is a Claude Code share link enough for a team? It is enough to show someone what happened. It is read-only, so it does not let a teammate steer or continue the work. Does a teammate need to install anything or sign up? To watch, neither: a viewer link opens in the browser. To steer, they sign in and join by editor link. Only the person whose machine runs the agent needs Claude Code installed. Can we collaborate with Codex or Cursor the same way? Yes. Friday finds Claude Code, Codex and Cursor on your computer and runs them in the same session. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/orca # Friday Work vs Orca By [Arun Dass](/about#arun) · Published 21 October 2026 Both run many coding agents at once. Orca is a free, open source development environment built for one person, with a git worktree for each task. Friday Work runs your agents in one live session that your team can watch, steer and take over, and the agents message each other. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## How do Friday and Orca compare? Friday Work Orca (from its own site) Price Free for one person; $15 or $30 a seat a month for teams Free and open source (MIT) Platforms In the browser, plus a desktop app for Windows and Mac Desktop apps for Mac, Windows and Linux, plus a mobile companion app Coding agents Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent 27 agents Keeping agents apart Each agent in its own lane, each in its own worktree or in Plan, with a warning when lanes in separate copies change the same file A git worktree for each task Connected tools 80+ over MCP GitHub and Linear Working as a team One live session: watch, steer, take over, hand off Built for one person today; no multi-user collaboration described Agents sharing context Agents message each other, ask questions and share notes Not described in what we checked Last checked: 21 October 2026 Based on Orca’s own site, checked 21 October 2026. Check onorca.dev for its current features. ## When is Orca the better pick? If you work alone, want everything free and open source, and want a worktree per task created for you, Orca is built for exactly that. It supports more coding agents than Friday, runs natively on Linux, and has a mobile companion app. If nobody else needs to see your agents, you gain little by adding a shared session. ## When is Friday the better pick? If a teammate needs to watch or steer an agent, if you want to hand a half-finished task to someone with its context, or if you want Claude Code and Codex to ask each other questions instead of you relaying them, Friday fits. Your own coding agents still run on your computer, under your own accounts, and Friday never meters them. ## Can you use both? Nothing stops you running a worktree-based tool for your own experiments and Friday for the work your team shares. For how parallel agents stay apart on one repository, see [how to run coding agents in parallel](/guides/run-coding-agents-in-parallel). ## Common questions Is Orca free? Yes. Its site describes it as free and open source under the MIT license. Does Orca support teams? Its site describes a tool for one person running many agents, and no multi-user collaboration. Check onorca.dev for anything newer. Does Friday use git worktrees? Yes, when you choose it: an agent can work in its own worktree, which Friday makes on your machine. Otherwise only one agent at a time can change the folder you chose; a second joins in Plan or gets its own worktree, and Friday warns you when lanes in separate copies change the same file. Which has more coding agents? Orca lists 27. Friday runs 14, plus its own agent, and any agent that speaks ACP. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/synara # Friday Work vs Synara By [Arun Dass](/about#arun) · Published 21 October 2026 Both keep your coding agents on your own computer. Synara is a free, local-first desktop app for one person running many agents, with handoff between providers. Friday Work puts a team and its agents in one live session, where people watch, steer and take over, and the agents message each other. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## How do Friday and Synara compare? Friday Work Synara (from its own site) Price Free for one person; $15 or $30 a seat a month for teams Free and open source (MIT) Platforms In the browser, plus a desktop app for Windows and Mac Desktop app for Mac, Windows and Linux Where agents run On your own computer, under your own accounts Local-first, on your own computer Switching agents Hand a task to another agent or a teammate, with the whole record Handoff between providers Working as a team One live session: watch, steer, take over Built for one person today; no sharing described Record A replayable record of every step, with approvals Not described in what we checked Last checked: 21 October 2026 Based on Synara’s own site, checked 21 October 2026. Check trysynara.com for its current features. ## When is Synara the better pick? If you are one developer who wants a free, local app to run several agents and move work between providers, and you have no need for anyone else to see it, Synara is simple and costs nothing. ## When is Friday the better pick? If a second person needs to see what an agent is doing, if work stalls when the person driving an agent is away, or if you want agents to share context with each other, Friday fits. A teammate joins by link in a browser, and can take over with the history intact. Friday also asks a person before an agent runs a command in Manual mode, in front of the whole session. Friday does store the session record so that your team can watch and replay it. The security page lists what is stored and what is not done yet. A purely local tool avoids that, at the price of not being shareable. ## How does handoff differ? In both, the point is to move work between agents without starting over. In Friday the handoff carries the task, what was found, which files were read and what was run, and it can go to a teammate as well as an agent. We measured the effect on one pair of agents in the [hand-off benchmark](/benchmarks/handoff), and it has limits that page lists. ## Common questions Is Synara free? Its site describes it as free and open source under the MIT license, as a local-first desktop app. Can I share a Synara session with a teammate? Its site describes a tool for one person, and no sharing. Check trysynara.com for anything newer. Does Friday work on Linux? Friday works in a browser on Linux. Its optional desktop app is for Windows and Mac. Do I need to pay to use Friday alone? No. Solo is free for one person, with your own coding agents unlimited and never metered. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/compare/orchestrator # Friday Work vs Orchestrator By [Arun Dass](/about#arun) · Published 21 October 2026 Both put a lead agent in charge of other agents. Orchestrator is an open source desktop app and CLI where an orchestrator agent spawns workers in git worktrees, built for one person, with collaboration on cloud agents announced for its Enterprise plan. Friday Work shares a live session with your team now, and its agents message each other. ## What is Friday Work? Friday Work is a multiplayer AI workspace for coding teams. Your people and your coding agents work in one live session: the same task, the same record, the same tools. It works in your browser; the desktop app for Windows and Mac is optional. - Teammates join a session by link, watch every step, suggest changes and take control when it is their turn. - Claude Code, Codex, Cursor and 11 more coding agents run on your own machine and your own plans, side by side in one session, with Friday’s own agent when you want it. - Smart routing suggests the agent and model most likely to get each task right, for the least, and a project brain briefs whichever agent comes next. - Continue a Claude Code, Codex or Cursor session you already have, instead of starting over. - Plan, Manual or Auto: you decide what an agent may do, and approve its commands as a team. - Connect GitHub, Slack, Linear and 80+ more tools over MCP once, and choose which agents may use them. - Free for one person. Team Basic is $15 a seat a month and Team Pro $30, and new teams get Team Basic free for 7 days. ## How do Friday and Orchestrator compare? Friday Work Orchestrator (from its own site) Main idea People and agents in one live session; any agent can lead and split a goal into a task board An orchestrator agent spawns worker agents and tracks them on a board Keeping agents apart Lanes, each in its own worktree or in Plan, with a warning when lanes in separate copies change the same file Workers in git worktrees Checking the work On a task board, a free agent reviews a finished task; a receipt lists the commands run and whether they passed CI feedback on the work Coding agents Claude Code, Codex, Cursor and 11 more, plus Friday’s own agent Dozens of agents Platforms In the browser, plus a desktop app for Windows and Mac Desktop app and CLI for Mac, Windows and Linux, plus a mobile companion app Working as a team Teammates watch, steer, take over and hand off today Collaboration on cloud agents, SSO and team management listed as coming soon for Enterprise, not shipped Price Free for one person; $15 or $30 a seat a month for teams Local app free and open source (Apache-2.0); cloud agents from $20 a month Last checked: 21 October 2026 Based on Orchestrator’s own site, checked 21 October 2026. Check orchestrator.inc for its current features. ## How are the two ideas alike? Both treat one agent as the planner. You give it a goal, it breaks the goal into pieces, and other agents take the pieces. In Friday a task names the files it owns and what it waits on, ready tasks go to free agents, and another agent checks a finished task before it counts. The difference is who else is in the room. ## When is Orchestrator the better pick? If you work alone, want a worktree per worker, want CI results fed back to the agents, and want an open source tool with more agents supported, Orchestrator is built for that. Its mobile companion app and CLI suit a person who manages work from several places, and it also offers paid cloud agents that keep running when your computer sleeps. Its collaboration features are announced for Enterprise, so check its site if you plan to wait for them. ## When is Friday the better pick? If teammates need to see and steer the work today, if you want a person to approve risky commands in front of the team (see [Claude Code permissions for teams](/guides/claude-code-permissions-for-teams)), or if you want agents that message each other and share notes, Friday fits. Your agents run on your own computer and plans, and Friday never meters them. ## Common questions Does Orchestrator have team features? Its site lists collaboration on cloud agents, SSO and team management as coming soon for Enterprise. Check orchestrator.inc for the latest. Is Orchestrator open source? Yes. Its site describes the local desktop app and CLI as free and open source under Apache-2.0. Paid plans add cloud agents. Can Friday run an orchestrator agent? Yes. Make any agent in a session the lead and give it a goal. It splits the work into a task board and hands tasks to the agents best placed for them. How does Friday check an agent’s work? Each run ends with a receipt listing the commands the agent ran and whether they passed, and on a task board a free agent reviews a finished task before it counts. A person still reads the diff and accepts it. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/share-codex-session-with-a-teammate # How to share a Codex session with a teammate By [Arun Dass](/about#arun) · Published 21 October 2026 · Updated 22 October 2026 Codex keeps your session on your own computer, so a teammate cannot join it. To share one, link your computer to Friday, continue your Codex session there, and send your teammate a link. They watch each step in a browser without signing up. A teammate who signs in as an editor can also suggest redirects and take control when it is their turn. Your Codex stays on your own plan. ## Can you share a Codex session? Codex is built around one person working with one agent. The usual ways to show a teammate what it did are a screen share, a pasted transcript, or a branch and pull request. Each gives them a view of the result, but the teammate cannot act inside your session. OpenAI’s documentation describes sharing plugins and environments with a workspace, not a running thread. We found nothing in Codex’s documentation that lets a second person join one live, so check OpenAI’s docs for anything newer. The common failure is the paste. A teammate who starts a fresh Codex session from your notes re-reads the same files and retries what you already ruled out. ## How do you share a Codex session in Friday? Friday runs your own Codex on your computer and opens the session to your team. 1. Link your computer to Friday with one command, or let the desktop app do it. Friday finds Codex, Claude Code and Cursor on your machine. 2. Pick the Codex session you already have and continue it in the same folder, so Codex resumes with its own context and Friday shares its latest messages with the session. 3. Choose what Codex may do: plan only, ask before each command, or run on its own (Plan, Manual and Auto). 4. Press Share in the session header and send the Watch live link. A viewer opens it in a browser with no account and no install. Send an editor link to someone who should steer. 5. Steer together. An editor suggests a redirect, the driver sends it to Codex, and anyone who can edit can ask for control. Codex runs under your own login and plan, and Friday never meters it. Your teammates do not need Codex at all unless they want to drive an agent from their own computer. See [how to continue a session with your team](/guides/continue-claude-code-session-with-your-team) for the import step in detail. ## What can your teammate do once they are in? Link type What the teammate can do Plan needed Viewer Read every step, including earlier messages and file changes, with no account Free on every plan Editor Sign in, then read, suggest redirects, ask for control, upload files Team, from $15 a seat a month The person driving Codex sees a teammate’s suggestion and decides whether to send it. When the driver is away, a teammate can take over and carry on with the whole session in front of them. Late joiners get a short digest of what changed while they were gone. ## What does sharing Codex add beyond a link? - Hand-off between tools. Move the task from Codex to Claude Code or Cursor with the whole record, or to a teammate. - Automatic hand-off at a limit. When a coding tool stops because it ran out of usage, Friday hands the task to another agent that can take it; if none can, it resumes when the limit resets or asks a person. - Smart routing. In Codex, Auto can pick the model for each message. - A record. Every step is kept, and anyone with access can replay it. For a measured look at what a hand-off saves, see the [hand-off benchmark](/benchmarks/handoff), which includes its limits. ## What should you know before you share? Your Codex keeps running on the computer you linked, so that computer has to stay on while it works. Friday stores the session record so that your team can read it, and anyone who can open the session can read all of it, earlier messages included. Treat the link as carefully as the work. The security page lists what is stored and what is not finished yet. Solo is free for one person, and anyone can watch for free. Teammates who steer with you need Team, with Team Basic at $15 a seat a month and a 7-day Team Basic trial for a new team, starting when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent. ## Common questions Can a teammate join my Codex session? Not in Codex on its own. In a Friday session they join by link, watch live, suggest redirects, and take control in turn. One person drives the agent at a time. Does my teammate need a Codex account? Only to run Codex from their own computer. To watch, they need no account at all, just the link. To suggest redirects or take control of your session, they sign in to Friday on Team. Can I share a Codex session I already started? Yes. Friday lists the Codex sessions on your linked computer so you can continue one, and shares its latest messages with the session. Does Friday charge for my Codex usage? No. Codex runs on your own plan and Friday never meters it. Solo is free for one person. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/share-cursor-session-with-a-teammate # How to share a Cursor session with a teammate By [Arun Dass](/about#arun) · Published 21 October 2026 · Updated 22 October 2026 Cursor’s editor chat lives in one person’s editor. Teams can share a read-only transcript, but a teammate cannot join it live. To share one live, run Cursor’s CLI agent in a Friday session on your own computer and send your teammate a link. They watch each step in a browser without signing up, and a teammate who signs in as an editor can steer with you. Friday shares the agent’s session, not your editor window. ## Can you share a Cursor session? Cursor is an editor with an agent inside it. Its Teams plan includes a marketplace for shared rules, skills and plugins, usage analytics, and shared transcripts: a Share button in the chat gives teammates a read-only copy of a conversation, tool calls included, and they can fork it into their own editor to continue. That is useful for review and for starting from someone’s context. It is a copy, not a live session, so two people still cannot work in one agent conversation as it runs. Cursor’s help page says shared transcripts are on Teams and Enterprise, not Pro, so check cursor.com for anything newer. So teams fall back on the same four things as with any coding agent: a screen share, pasted output, a branch, or a call. Our guide to Claude Code collaboration compares them, and the tradeoffs are the same for Cursor. ## How do you share a Cursor session in Friday? 1. Link your computer to Friday with one command, or let the desktop app do it. Friday finds Cursor, Codex and Claude Code on your machine. 2. Run Cursor’s CLI agent in a Friday session, or continue a session you started with the Cursor CLI. Chats in the Cursor editor cannot be continued in Friday. 3. Set the mode: Plan, Manual or Auto. 4. Press Share in the session header and send the Watch live link. A viewer opens it with no account. Send an editor link to someone who should steer. 5. Steer together in the browser. Editors’ suggestions go to the driver, and anyone who can edit can ask for control. Cursor runs on your computer under your own login and plan, and Friday never meters it. ## What does a teammate see that they could not before? Cursor on its own Cursor in a Friday session Who sees the agent’s steps Whoever is at your editor Everyone with the link, live, in a browser, no account needed Who can redirect it You Anyone who suggests; the driver sends it Handing it to someone else Describe the state in a message Pass control with the full session Record afterward Your own chat history A session record anyone with access can replay Teammates who use a different tool are fine. In one session your Cursor agent can sit beside Claude Code and Codex, and work can move between them with its history. ## What are the limits? - Friday shares the agent session. It does not stream your Cursor editor, so a teammate sees the agent’s steps and changes, not your open tabs. - The agent runs on the computer you linked, which has to stay on while it works. - Cursor can use Friday’s agent-to-agent tools only in Auto mode. In Manual and Plan it works as a normal agent in the session, but it will not message the others. - Anyone who can open the session can read all of it, earlier messages included. Treat the link as carefully as the work. For how agents in one session ask each other questions, see [Claude Code and Codex talking to each other](/guides/claude-code-and-codex-talk-to-each-other). ## What does it cost? Solo is free for one person, with your own Cursor never metered, and anyone can watch for free. Teammates who steer with you need Team: Team Basic is $15 a seat a month, Team Pro $30, and a new team workspace gets Team Basic free for 7 days, starting when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent. No card, one trial per account. ## Common questions Can two people use one Cursor agent chat? Not in Cursor on its own. In a Friday session teammates watch live, suggest redirects and take control in turn, one driver at a time. Does my teammate need Cursor? No. To watch they need no account and no install, only the link. To steer, they sign in to Friday on Team. They need Cursor installed only if they want to run their own Cursor agent. Does Friday replace Cursor? No. Your own Cursor runs on your own plan. Friday adds the shared session around the agent. Can I move work from Cursor to another agent? Yes. A hand-off carries the task and the record to Claude Code, Codex or another agent, or to a teammate. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/ai-coding-agent-orchestrator # AI coding agent orchestrators: what they are and how to choose one By [Arun Dass](/about#arun) · Published 21 October 2026 An agent orchestrator coordinates several coding agents so they work on one goal without colliding: it splits the work, gives each agent its own space, tracks progress and checks results. Some orchestrators are apps for one person. Others are shared with a team. Pick by who needs to see the work and where the agents run. ## What is an agent orchestrator? A coding agent on its own takes a task and does it. Run three of them and you now have a coordination problem: who does what, who is touching which files, which one is stuck. An orchestrator takes that job off you. The word covers two things that are often combined. One is an app that manages many agents for you, creating a worktree per task and showing every agent’s status. The other is a lead agent that plans: you give it a goal, it breaks the goal into tasks and hands them to other agents. ## What does an orchestrator do? - Splits a goal into tasks that do not share files. - Gives each agent its own working copy, usually a git worktree, so edits cannot collide. - Shows every agent’s status in one place. - Passes context between agents, so each one does not start cold. - Checks finished work, with tests, a reviewing agent or a person. - Moves a task to another agent when one is stuck or out of usage. None of this removes the review. Three agents still produce three sets of changes, and [reviewing AI-generated pull requests](/guides/review-ai-generated-pull-requests) is still your job. ## How do you choose one? Question Why it matters Is it for one person or a team? Several tools are built for one user today. If a teammate must see or steer the work, check what is shipped, not announced. Where do the agents run? On your computer, under your own accounts, or on the vendor’s cloud machines. Cloud keeps working when your laptop sleeps; local keeps work on your own machine and plans. Which coding agents does it run? Counts vary from a handful to dozens. What matters is whether it runs the ones you already pay for. How does it keep agents apart? A worktree per task, a warning when two agents change one file, or both. How is work checked? Tests, a reviewing agent, or only you. What does it cost and who owns the code? Free and open source tools exist. Paid ones charge per seat or for cloud agents. ## Which orchestrators exist? Tool Built for Worth knowing (from each product’s own site, 21 October 2026) Orca One person Free, open source (MIT), desktop apps for Mac, Windows and Linux, a worktree per task, 27 agents Synara One person Free, local-first desktop app, handoff between providers Orchestrator One person today Open source (Apache-2.0), an orchestrator agent spawns workers in worktrees, paid cloud agents; collaboration listed as coming for Enterprise Conductor One person, and teams on paid plans Parallel agents in isolated workspaces; paid plans add cloud workspaces and multiplayer; local app is Mac only AQ Teams Browser workspace; agents run on a VM, never on your laptop Friday Work One person, and teams Your own agents on your own computer in one shared session; any agent can lead and split a goal into a task board We compare four of these in detail: [Orca](/compare/orca), [Synara](/compare/synara), [Orchestrator](/compare/orchestrator) and [Conductor](/compare/conductor). Check each site before you decide, since features change monthly. ## Where does Friday Work fit? Friday is a free agent orchestrator for your coding tools, with Team for working on it together. On Solo you run Claude Code, Codex, Cursor and 11 more agents side by side on your own plans, with smart routing and a project brain for each repository. Any agent can be made the lead: give it one goal and it posts a task board, and free agents pick tasks up. Agents message each other and share notes. Where it is the wrong pick: if you want desktop apps for Mac, Windows and Linux, a mobile app, or the largest list of supported agents, Orca covers more. Friday’s desktop app is for Windows and Mac, and it works in a browser anywhere. Friday earns its place when a second person needs to see and steer the agents, or when you want agents that talk to each other. ## How do you start? 1. Start with one real task that splits into two or three pieces, such as an endpoint, its tests and its docs. 2. Try one orchestrator at a time on your own repository for a few days. 3. Notice where you still do the coordinating by hand. That is the feature to look for. 4. Add a second person only when you need it. The next guide, [a multi-agent coding workflow](/guides/multi-agent-coding-workflow), walks through a lead agent and a task board. ## Common questions What is an AI coding agent orchestrator? A tool or lead agent that coordinates several coding agents on one goal: it splits the work, keeps agents apart, tracks progress and checks results. Is there a free agent orchestrator? Yes. Orca and Synara are free and open source for one person, and Friday Work is free for one person, with your own agents never metered. Do I need an orchestrator for two agents? Not always. Two agents on separate folders may need only discipline. An orchestrator pays off once you run three or more, or share the work. Can an orchestrator run Claude Code and Codex together? Several can. Friday and Conductor run both, and the others list many agents, so check each site for the ones you use. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/multi-agent-coding-workflow # A multi-agent coding workflow: one lead agent and a task board By [Arun Dass](/about#arun) · Published 21 October 2026 In a multi-agent coding workflow, one agent plans and the others build. You give the lead a goal. It posts a task board, with each task naming the files it owns and what it waits on. Free agents take ready tasks, and a different agent checks each finished task before it counts. This is how it works in Friday. ## What is the shape of the workflow? 1. Add the agents you want to the session, such as Claude Code, Codex and Cursor. 2. Make one the lead from its tab’s menu, with “Make it lead the others”. An owner or editor sets it, and only one agent leads at a time. 3. Give the lead one goal, in the All agents view or its own lane. 4. The lead splits the work and posts the board. 5. Free agents take ready tasks. Another agent reviews each finished task. 6. The lead hears once, when the board is through, and reports what was done. ## What goes on the task board? Each task has a title, what to do, the files it owns, the tasks it waits on and the agent it suits. The lead is told to cut work into pieces that do not share files and to hand out every piece up front, so they run at once, and to do a piece of its own meanwhile. Field What it does Files it owns Keeps two agents from editing the same file Waits on Holds a task until the ones it depends on are done Suits Names the agent best placed for it, which gets first call Status Shown live on one card: who has each task, who is checking it A task is ready when everything it waits on is done and nobody has it. A busy agent picks up the next ready task by itself when it finishes, so nobody dispatches by hand. ## How does the work get checked? A finished task goes to a free agent that did not do it. That agent reads the change, runs the tests and ends with “Approved” or “Changes needed”. If approved, the task is done and what waited on it can start. If not, it goes back to its agent, and after a third failed review it is blocked for the lead or a person. Each report ends with facts from the session record rather than the agent’s own account: files changed with additions and deletions, the tests and checks run and how they came out, any commit, and how long it took. A task that changed no files, or with no other agent to check it, passes with a note. ## How do the agents stay in step? - A task names the other open tasks and who has them, so agents settle what they share, such as an interface, by asking each other. - Agents say which files they will change before they write, and Friday warns when lanes in separate copies change the same one. - An agent that stops short of its task is asked to carry on, up to five times in a run, unless it is waiting on a person. - When a coding tool runs out of usage, its task is handed on to another agent that can take it, or resumes when the limit resets. - A failed or blocked task reaches the lead at once, so it can unblock it while the others carry on. The messaging underneath is covered in [Claude Code and Codex talking to each other](/guides/claude-code-and-codex-talk-to-each-other). ## Where do people come in? The board is not unattended automation. Everyone in the session sees the board and every agent’s steps, and a teammate can suggest a redirect or take control of an agent. Plan, Manual and Auto still decide what each agent may do without asking, and risky commands wait for a yes. A person reads the diff and accepts it. Friday ranks each run’s changes by risk, so the small safe ones can be cleared together. ## When is this too much? For a change that fits in one agent’s head, a lead and a board is overhead. Use it when a goal splits into three or more pieces that touch different files. Friday allows one writer per folder on a computer, so give each agent that edits its own worktree. Lanes in separate copies still get a warning when they touch the same file. Your usage limits matter as well: several agents on one plan reach the limit sooner than one. If you are still choosing a tool for this, see [how to choose an agent orchestrator](/guides/ai-coding-agent-orchestrator). ## Common questions Can any agent be the lead? Yes. Any agent in the session can be made the lead from its tab’s menu, one at a time. Who checks the agents’ work? Another agent checks each finished task against its report and the tests, and a person still reads the diff and accepts it. Do I need to dispatch tasks myself? No. Ready tasks go to free agents on their own, and busy agents take the next ready one when they finish. Does this cost extra? Your own Claude Code, Codex and Cursor run on your own plans and Friday never meters them. Solo is free for one person. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/watch-a-teammates-coding-agent-live # How to watch a teammate’s coding agent live By [Arun Dass](/about#arun) · Published 21 October 2026 · Updated 22 October 2026 To watch a teammate’s coding agent live, they send you a link to their Friday session. You open it in a browser and see every step as it happens: messages, tool calls and file changes. You do not need an account, and watching is free on every plan. You can read earlier steps too, and replay the session later. ## Why would you watch someone else’s agent? - To catch a wrong turn before it costs an hour. - To learn how a stronger engineer prompts and steers. - To follow an engineer’s onboarding by watching real sessions. - To see where a teammate’s work stands without asking. - To prepare for taking the session over. A teammate’s agent normally lives in their terminal, where nobody else can see it. Screenshots and a call only cover the moments you are both there. ## How do you watch? 1. The teammate runs their Claude Code, Codex or Cursor in a Friday session on their own computer. 2. They press Share in the session header and send you the Watch live link. The same menu has a Replay link, and links show a preview when pasted into chat. 3. You open the link in a browser. There is no account to make and nothing to install. 4. You see the session as it runs, read-only. A short digest says what changed if you arrive late. A slim bar on the page offers Try Friday free, if you want a session of your own. The desktop app for Windows and Mac is optional. ## What can a viewer see and do? Viewer Editor Account needed No Yes, sign in Read every step, including tool calls and file changes Yes Yes Read what happened before joining Yes Yes Suggest a redirect to the driver No Yes Ask for control of the agent No Yes Cost Free on every plan Team, from $15 a seat a month The session owner can change what editors may do. To move from watching to steering, ask for an editor link and sign in. ## Can you watch a run later? Yes. Every session keeps a record of each step, and you can wind it back to any moment. That helps when something broke overnight, or when a new teammate wants to see how a feature was built. How long the record is kept depends on the plan: history is kept for at least 30 days on Solo and at least 365 on Team Basic. Nothing deletes older sessions automatically yet. ## What should the person sharing keep in mind? Anyone who can open the session can read all of it, including earlier messages, so treat the link as carefully as the work. The agent keeps running on the computer it was linked on, so that machine has to stay on while it works. The session record is stored by Friday so your team can read it. The security page says what is stored and what is not finished yet. If the teammate wants your help and not just your eyes, see [Claude Code collaboration](/guides/claude-code-collaboration), and for passing the work on, [hand off a coding agent session](/guides/ai-agent-handoff). ## Common questions Is watching a coding agent free? Yes. Watching a session through a viewer link costs nothing on every Friday plan. Steering with the person needs Team. Does the viewer need an account or to install anything? No to both. A viewer link opens in a browser and shows the session read-only. Steering needs an editor link, an account and Team. The desktop app is optional. Can a viewer see what the agent did before they joined? Yes. Anyone who can open the session can read it from the start, and a late joiner also gets a short digest. Can I replay a session afterward? Yes. Each session keeps a record you can replay, within the history your plan keeps. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/send-a-link-to-watch-an-ai-coding-agent # How to send a link so someone can watch an AI coding agent run By [Arun Dass](/about#arun) · Published 22 October 2026 To let someone watch an AI coding agent run, run it in Friday, press Share in the session header and send the Watch live link. The person opens it in a browser and sees every step as it happens. They do not need an account, and it costs nothing. They can only read; steering takes an editor link, an account and Team. ## Who is this for? - A teammate who wants to see where your agent is, without a call. - A manager, client or contractor who has no Friday account and should not need one. - A friend you want to show a run to, or someone helping you debug it. - Anyone you would otherwise send a screen recording or a pasted transcript. ## How do you send the link? 1. Run your Claude Code, Codex, Cursor or another coding agent in a Friday session on your own computer. 2. Press Share in the session header. 3. Copy the Watch live link, or the Replay link if the run has finished. 4. Paste it into chat, email or an issue. It shows a preview with the session’s name. The agent keeps running on your computer, so that computer has to stay on while it works. ## What does the person see? They see the session as it runs, in a browser: messages, tool calls and file changes, and what happened before they opened it. The page is read-only. A slim bar offers Try Friday free, if they want a session of their own. Watch live link Editor link Account needed No Yes, they sign in Can read the whole session Yes Yes Can suggest a redirect or ask for control No Yes Cost Free on every plan Team, from $15 a seat a month ## What should you check before you send it? - Anyone with the link can read the whole session, including earlier messages and anything an agent printed. Keep secrets out of the folders agents work in. - Share only with people who should see the work. You can turn a link off or replace it at any time. - Friday stores the session record so the link works. The security page says what is stored and what is not finished yet. For the same flow written for a teammate on the other end, see [how to watch a teammate’s coding agent live](/guides/watch-a-teammates-coding-agent-live). To let them act as well, see [how to share a Claude Code session with your team](/guides/share-claude-code-session-with-your-team). ## Common questions Does the person need a Friday account to watch? No. A Watch live link opens in a browser with no account and no install. The page is read-only. Does it cost anything? No. Watching is free on every plan and never needs a seat. Solo is free for one person. Can they type to the agent? Not from a viewer link. To steer, send an editor link: they sign in, and steering needs Team. What is the difference between Watch live and Replay? Watch live follows a session as it runs. Replay opens the recorded session after the fact. Does it work with any coding agent? Any agent you run in a Friday session, including Claude Code, Codex and Cursor, on your own plan. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/guides/replay-a-coding-agent-session # How to replay a coding agent session By [Arun Dass](/about#arun) · Published 22 October 2026 To replay a coding agent session, run the agent in Friday, which records every step, then open the session or send its Replay link. The person who opens it can read what the agent did from the start, and they do not need an account. Plans keep the record for 30 days on Solo and 365 on Team. ## Why replay a session? - To find out what an agent did overnight, and where it went wrong. - To show a teammate how a feature was built, step by step. - To review a change by reading the run behind it, not only the diff. - To keep a record of who asked for what, and what the agent ran. A chat transcript in one agent’s terminal ends when the terminal closes, and it stays on one machine. A replay keeps the whole run in one place that a link can open. ## How do you replay one? 1. Run your coding agent in a Friday session. Every step is recorded as it happens. 2. Open the session to wind it back to any moment that mattered: an instruction, a change, a commit or a change of driver. 3. To let someone else watch, press Share in the session header and send the Replay link. The Replay link opens in a browser, with no account and nothing to install. It is read-only. ## What is kept in the record? Every action is logged with the person who asked for it, what the agent did, what each tool returned and when. After a run, a receipt lists the commands it really ran and whether they passed. Solo Team Basic Team Pro How long the record is kept At least 30 days At least 365 days At least 365 days Older sessions are not removed automatically yet. ## What should you know before you share a replay? Anyone with the link can read the whole record, including anything an agent printed, so keep secrets out of the folders agents work in and share only with people who should see the work. Friday stores the session record so a replay can be opened. The security page says what is stored and what is not finished yet. To watch a session while it runs, see [how to send a link so someone can watch an AI coding agent run](/guides/send-a-link-to-watch-an-ai-coding-agent). For what a record should hold, see [what is an audit trail for AI agents](/guides/ai-agent-audit-trail). ## Common questions Can someone replay a session without a Friday account? Yes. A Replay link opens in a browser with no account. It is read-only. How long can I replay a session? For at least as long as your plan promises: 30 days on Solo and 365 days on Team Basic and Team Pro. Does replay work for Claude Code, Codex and Cursor? For any coding agent you run in a Friday session, since Friday records the session. Is a replay the same as an exported transcript? No. A transcript is a static file. A replay is the recorded session itself, which you can wind back to any moment that mattered. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/docs # Friday Work documentation How to set up Friday, bring your team in, and work with agents. [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) ## Start - [**Getting started**Create an account, make a workspace, start a shared session with an agent and invite your team.](/docs/getting-started) - [**Your coding agents**Use your own Claude Code, Codex or Cursor in a Friday session, choose what the agent may do on your computer, and add the Friday plugin to Claude Code.](/docs/coding-agents) - [**Invite your team**Share a session by editor or viewer link, see who can read what, and check the limits for each plan.](/docs/invite-your-team) ## Use - [**Agents and skills**Describe what an agent should do and the agent builder creates it. Add skills, set it to run on a schedule, and share it with your team.](/docs/agents-and-skills) - [**Control and limits**How the driver works, what everyone else can do, how to review a change and what the session record keeps.](/docs/control-and-limits) - [**Connect tools**Connect 80+ tools over MCP, choose which ones agents may use and see every call in the session feed.](/docs/connect-tools) ## Plans - [**Plans and credits**What Solo, Team Basic and Team Pro include, how Team Pro’s frontier credits work in 5-hour windows and what happens when you run out.](/docs/plans-and-credits) --- Source: https://www.tryfridaywork.com/docs/getting-started # Getting started Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Create an account, make a workspace, start a session and bring your team in. This takes a few minutes. ## What you need An account, and one task you want to work on with an agent. To use your own Claude Code, Codex or Cursor, you also need them installed on your computer. ## Set up your first session 1. Create an account. You can sign in with Google, with GitHub, or with an email address and password. 2. Open your workspace. Work lives in a workspace, and its owner decides who is a member, which plan it runs on, and which connections agents can use. 3. Start a session for the task. A session holds one or more agents. 4. Add an agent. Use Friday’s own, one your team has built, or your own coding agent. 5. Invite your team by link, as editors or viewers. ## What happens in a session Everyone in the session sees the same messages, files and tool calls as they happen. Each agent has one driver at a time, and everyone else can watch, suggest a redirect or ask to take over. Every action is recorded in the session log. See [control and limits](/docs/control-and-limits). ## Where to go next - [Use your own coding agents](/docs/coding-agents): Claude Code, Codex or Cursor. - [Invite your team](/docs/invite-your-team) and choose what they can do. - [Build agents and skills](/docs/agents-and-skills) your team can reuse. ## Related - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) - [Plans and credits](/docs/plans-and-credits) --- Source: https://www.tryfridaywork.com/docs/coding-agents # Use your own coding agents Updated 23 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Run your own Claude Code, Codex or Cursor in a Friday session. They run on your machine, under your own account, and Friday never meters them. ## Link your machine In the desktop app, your machine links automatically. Friday finds the coding agents installed on it: Claude Code, Codex and Cursor. In a browser, Friday cannot see your computer, so you link it once. The Code tab shows a one-line command to run on your machine, or on Windows a file you double-click. Leave it running and your agents appear in Friday. ## Pick the code to work on Choose who does the work, then which code. You can search your recent repositories and your GitHub repositories, paste a repository URL, or drop in a folder. The sessions your coding agent already has are listed too. Pick one and Friday continues that conversation, with its context and memory. Work lands on a new branch named after what you asked for. ## Choose what an agent may do The host chooses what the agent may do unattended. Friday never goes beyond the mode you pick. Mode What the agent can do Plan only Read and plan, without changing anything Edit files Change files in the folder you choose. Shell commands are refused. Edit and run commands Change files and run commands in that folder Within a mode, an agent can change or delete files and run commands with real effects. Point it at a folder under version control, keep your own backups, and do not link a machine that holds anything you cannot afford to lose. ## Cost Your own coding agents run under your own subscription with their provider. Friday does not hold your credentials for them and does not meter them on any plan. ## Use Friday from Claude Code Friday has a plugin for Claude Code. Inside Claude Code, add it and install it, then run `/mcp` and pick `friday` to sign in (OAuth): ``` /plugin marketplace add splashxmoon/friday-claude-plugin /plugin install friday@friday-work /mcp ``` You get Friday’s MCP server, which can read your Friday profile and sessions and post a message into a session as you, and the command `/friday:share`, which opens Friday in your browser on the current folder and Claude Code session, to continue it there with your team. The plugin is [on GitHub](https://github.com/splashxmoon/friday-claude-plugin), and the server is listed in the [official MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers/com.tryfridaywork%2Ffriday-work/versions/latest) as `com.tryfridaywork/friday-work`. ## Related - [Control and limits](/docs/control-and-limits) - [Getting started](/docs/getting-started) - [How to share a Claude Code session with your team](/guides/share-claude-code-session-with-your-team) --- Source: https://www.tryfridaywork.com/docs/invite-your-team # Invite your team Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Share a session by link as an editor or a viewer. In the desktop app you can also invite a person by name. ## Editor and viewer links Role By default, the person can Editor Read the session, suggest changes, ask for control and upload files Viewer Read the session, without an account The session owner can change what editors and viewers are allowed to do: suggest, take control, upload and invite. Anyone with a viewer link can watch without an account; the page is read-only and offers a way to try Friday free. Steering takes an editor link, an account and Team. Treat links as carefully as the work itself. ## Invite a person by name In the desktop app there is no address bar, so you can pick a name from the workspace directory instead. The invite arrives in that person’s sidebar, and they accept or decline it. ## Who can see what Anyone who can open a session can read all of it, including what happened before they joined. A session is a room with a transcript, not a private message. If your workspace belongs to an employer or a client, that organization controls its content and can add or remove members. ## Limits by plan Solo Team Basic Team Pro People working in a session Just you (anyone can watch) Up to 25 Up to 25 Workspace members 1 Unlimited Unlimited ## Helping someone catch up Anyone who joins late gets a short digest of what changed while they were away, built from the session log. They can read the full session from the start. ## Related - [Control and limits](/docs/control-and-limits) - [Plans and credits](/docs/plans-and-credits) - [Shared sessions](/shared-sessions) --- Source: https://www.tryfridaywork.com/docs/agents-and-skills # Agents and skills Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) An agent is a worker with its own instructions, skills and model. Describe what it should do, and the agent builder creates it. ## Build an agent You build agents in the agent builder. You describe what the agent should do, and the builder creates it. 1. Open the agent builder. 2. Type what the agent should do, in your own words. 3. The builder creates the agent, with its instructions, tools and skills. It can also add helpers and choose skills that fit the job. 4. Review what it made, edit anything you want, and choose a model. 5. Share it with named teammates or the whole workspace. You can also start from a template, or save your own for next time. ## What an agent can have - Instructions: its brief. - Tools: web search, asking a person, and the connections it may use. - Skills: briefs for particular jobs, which the agent reaches for when a job matches. - Memory: facts it should always know. - Helpers: small specialists written into the agent, which it hands part of a job to. - A channel: where it reports when it finishes, in a connected tool such as Slack. - A trigger: a webhook URL that starts a session running this agent, so another tool can set it off. You can set a schedule in the builder, but scheduled runs are not live yet. ## Skills A skill is a brief you write once, such as “Review a diff”. Any agent that carries it gets better at that job. A skill has keywords that say when it applies, and can carry reference files. You can also import skills from a GitHub repository. Paste its address, pick the skills you want, and add them to your workspace. ## Memory Friday can remember what should still be true next week, for you and for your workspace. You decide whether memory is read into your sessions and whether agents may save to it. You can see, add and delete every entry. ## Models Friday’s own agent runs on Friday’s models on the Team plans: the default model on Team Basic, and frontier models on Team Pro, from Anthropic, OpenAI, Google and DeepSeek. On any plan, including Solo, it can run on your own model instead: any API key, or a model running on your computer, never metered. You choose a model for each agent, and a thinking level: fast, balanced or deep. ## Several agents in one session Each agent works in its own lane. A lane can wake when another lane finishes, which is how you build a pipeline, and you can save a pipeline to reuse it. Solo Team Basic Team Pro Extra Friday agents in a session None 10 10 Your own coding tools per session Unlimited Unlimited Unlimited Handoffs between agents Yes Yes Yes ## Related - [Agent workspace](/agent-workspace) - [Plans and credits](/docs/plans-and-credits) - [Hand off a coding agent session to a teammate](/guides/ai-agent-handoff) --- Source: https://www.tryfridaywork.com/docs/control-and-limits # Control and limits Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Each agent has one driver at a time. Everyone else can watch, suggest and ask to take over, and the session owner can always pause or cancel. ## The driver The driver is the person whose instructions an agent follows. Each agent has its own driver, so one person can steer one agent while a teammate steers another. Only the driver can instruct the agent, interrupt it without restarting, and resume it after a pause. The driver or the session owner can pause or cancel a run. ## What everyone else can do - Watch every step as it happens. - Suggest a redirect. It never reaches the agent on its own. The driver sees it and can send it with one click. - Request control. In a session set to ask, the driver approves, so the log shows who approved each handoff. In a session set to take over, anyone allowed to drive can take control straight away. If the driver disconnects, control passes to the person who has been waiting longest, or is released. ## Permissions The owner can always do everything. For editors and viewers, the owner chooses whether they can suggest, take control, upload and invite. By default, editors can suggest, take control and upload, and viewers can only read. ## Plan first A session can be set so the agent may read and plan but changes nothing until the driver approves its plan. You can also require sign-off from a number of teammates first. After changing code, the agent finds the project’s tests and runs them, unless you turn that off. ## Review a change Each run comes with a receipt: the commands it really ran, whether they passed, and a flag when a test was weakened or the agent claimed a pass that never ran. You read the diff and accept it, or ask for changes. You can undo a run. Ship opens a dialog with a commit message already written from the diff, and can commit, push and open a pull request using your machine’s own git login. ## Files and overlaps On one computer, only one agent can change a given folder at a time: a second agent joins in Plan, or gets its own worktree. If lanes working in separate copies change the same file, Friday warns you and says who else changed it and when. ## The record and Replay Every action is logged with the person who asked for it, what the agent did, what each tool returned and when. Replay winds a session back to any moment that mattered: an instruction, a change, a commit or a change of driver. Solo Team Basic Team Pro Session history Kept at least 30 days Kept at least 365 days Kept at least 365 days ## Related - [Shared control of AI agents: the driver model](/guides/shared-control-of-ai-agents) - [Human in the loop for coding agents](/guides/human-in-the-loop-coding-agents) - [What is an audit trail for AI agents?](/guides/ai-agent-audit-trail) --- Source: https://www.tryfridaywork.com/docs/connect-tools # Connect tools over MCP Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Agents can use 80+ tools over MCP, including Google Workspace, Slack, GitHub and Stripe. A workspace admin adds the connections, and you choose which ones each agent may use. ## What is MCP? The Model Context Protocol is an open standard for connecting AI agents to tools and data. Friday has a catalog of connections to common tools. ## What you can connect - Google Workspace: Drive, Gmail, Calendar, Docs, Sheets, Slides and Chat, plus Microsoft 365. - Chat and docs: Slack, Notion, Box, Miro, Airtable, Fireflies and Canva. - Code and design: GitHub, GitLab, Figma, Postman, Hugging Face, DeepWiki, Context7, GitMCP and Microsoft Learn. - Planning: Linear, Atlassian Rovo, Asana, ClickUp and monday.com. - Support and sales: HubSpot, Intercom, Zendesk, Attio, Close and Zapier. - Payments: Stripe, PayPal, Square and Plaid. - Monitoring: Sentry, Datadog, PagerDuty and Grafana. - Hosting and data: Vercel, Netlify, Supabase, Neon, Prisma, Cloudflare, Webflow and Wix. See the full list on the [integrations page](/integrations). ## Good to know Asana, Intercom, Square and Vercel only let apps connect from approved addresses, so they connect from the desktop app, not from a browser. Figma, Grafana and Microsoft 365 run as small servers on your own machine. Some providers, including Google Workspace, Slack and GitHub, need an OAuth app registered for your Friday. The connection card says so when that applies. ## Access and safety A workspace admin adds and manages connections. Their tools are offered to sessions in the workspace. You can limit an agent to all connections, none, or a chosen few, and a session can switch connections off. An agent can use a connected tool only within the permissions you approved, including tools that change data. Connect only what the work needs. Every tool call shows in the session feed. A tool that changes data finishes its call once it starts. Only tools marked read-only can be stopped mid-call. ## Plans MCP connections are included on Solo, Team Basic and Team Pro. ## Related - [Integrations](/integrations) - [MCP for teams: connect AI agents to your tools](/guides/mcp-for-teams) - [Control and limits](/docs/control-and-limits) --- Source: https://www.tryfridaywork.com/docs/plans-and-credits # Plans and credits Updated 1 October 2026 [Overview](/docs) Start - [Getting started](/docs/getting-started) - [Your coding agents](/docs/coding-agents) - [Invite your team](/docs/invite-your-team) Use - [Agents and skills](/docs/agents-and-skills) - [Control and limits](/docs/control-and-limits) - [Connect tools](/docs/connect-tools) Plans - [Plans and credits](/docs/plans-and-credits) Solo is free. Team Basic and Team Pro are per seat. Team Pro’s frontier models are counted in credits, in 5-hour windows, and Friday’s default model and your own coding agents are never the limit. ## The plans Solo Team Basic Team Pro Price Free $15 a seat a month $30 a seat a month Paid yearly — $144 a seat ($12 a month) $300 a seat ($25 a month) Your own coding tools Unlimited, never metered Unlimited, never metered Unlimited, never metered Friday’s own agent On your own model Default model, unlimited Frontier models, refilled every 5 hours, plus the default model, unlimited People working in a session Just you (anyone can watch) Up to 25 Up to 25 Workspace members 1 Unlimited Unlimited Extra Friday agents in a session None 10 10 Session history Kept at least 30 days Kept at least 365 days Kept at least 365 days A new team gets Team Basic free for 7 days, without a card. The 7 days start when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent, and each account gets one trial. See the [pricing page](/pricing) for the full comparison. ## How the frontier allowance works Team Pro includes 550 credits of frontier models for Friday’s own agent in every 5-hour window. Bigger models use credits faster. A window starts at your first message and resets five hours later, and the meter in the app shows what is left and when it starts over. When the credits run out, Friday’s own agent carries on with the default model, which is unlimited. ## What if you run out? Your session keeps going. Switch Friday’s agent to its default model, which is unlimited on both Team plans, or keep working with your own Claude Code, Codex or Cursor. The frontier allowance starts over when the window ends. ## What is never metered? Your own Claude Code, Codex or Cursor, on every plan. ## Related - [Pricing](/pricing) - [Agents and skills](/docs/agents-and-skills) - [Your coding agents](/docs/coding-agents) --- Source: https://www.tryfridaywork.com/pricing # Free on your own. Per seat with your team. Run every coding agent side by side for free. Bring your team in from $15 a seat, and a new team gets Team Basic free for 7 days. ## Solo For one developer Freeforever [Try for free →](/signup) No card. Works in your browser. - Claude Code, Codex, Cursor and 11 more coding agents side by side, on your own plans, never metered - Smart routing and a project brain for every repo - Friday’s own agent on your own model: any API key, or one running on your computer - Import a session you already started and carry on - Anyone can watch your sessions with a link - History kept for at least 30 days ## Team Basic 7 days free For teams that build together $15a seat a month billed monthly [Try Team Basic free →](/signup) 7 days free. No card required. - Everything in Solo - Up to 25 people working in one session - Unlimited workspace members - Friday’s own agent on its default model, unlimited - 10 extra Friday agents in a session - History kept for at least 365 days ## Team Pro For teams that want frontier models $30a seat a month billed monthly [Choose Team Pro →](/signup?plan=pro&billing=monthly) Cancel anytime. - Everything in Team Basic - Frontier models for Friday’s own agent, refilled every 5 hours - Friday’s default model, still unlimited - **No card to start**Solo and the Team Basic trial - **Never metered**Your coding agents run on your own plans - **Watching is free**Anyone with a link, no seat needed - **Cancel anytime**Your plan runs to the end of the period ## Which plan is right for me? 1. **I code on my own.**Claude Code, Codex, Cursor and 11 more side by side, with smart routing and a project brain.[Solo →Free](/signup) 2. **My team wants to work in one live session.**Teammates watch, steer and take over, with Friday’s own agent on its default model. Start with 7 days free.[Team Basic →$15 a seat a month](/signup) 3. **We want frontier models too.**Everything in Team Basic, plus frontier models for Friday’s agent.[Team Pro →$30 a seat a month](/signup?plan=pro&billing=monthly) ## Compare the plans. Swipe to see every plan → Feature SoloFree Team Basic$15 a seat a month Team Pro$30 a seat a month Run your agents Your coding agents (Claude Code, Codex, Cursor and 11 more), on your own plans Included Included Included Friday’s own agent On your own model Default model, unlimited Frontier models, refilled every 5 hours, plus the default model, unlimited Your own models (any API key or a local model), never metered Included Included Included Smart routing Included Included Included Project brain Included Included Included Extra Friday agents in a session — 10 10 Security and control Your coding agents run on your computer, under your own login Included Included Included Plan, Manual or Auto for each agent, and approvals in front of everyone Included Included Included Roles: owner, editor and viewer You, and viewers Included Included Invite links you can turn off, and people removed at once Included Included Included Traffic encrypted in transit Included Included Included SOC 2, ISO 27001 or enterprise single sign-on — — — Work together People working in one session Just you (anyone can watch) Up to 25 Up to 25 Workspace members 1 Unlimited Unlimited Session history Kept at least 30 days Kept at least 365 days Kept at least 365 days ## Questions. What’s the difference between Solo and Team? Solo is for one person running their own coding agents side by side, and it’s free, with smart routing and a project brain. Bring your own model and Friday’s own agent works on Solo too. Team is for working together: teammates join the same live session to steer and take over, and Friday’s own agent runs on Friday’s models, the default model on Team Basic and frontier models on Team Pro. How does the trial work? A team gets Team Basic free for 7 days, without a card. The 7 days start when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent, and each account gets one trial. When the trial ends, choose Team Basic or Team Pro to keep working together, or carry on in Solo. What is a seat? A seat is one person on your team. Anyone can watch a session with a viewer link on any plan, without an account, and watching never needs a seat. Is there a yearly price? Yes. Paid a year up front, Team Basic is $12 a seat a month ($144 a year) and Team Pro is $25 ($300 a year). Is Friday for enterprise teams? No. Friday is built for individual developers and small teams working with coding agents. It has no SOC 2 or ISO 27001 and no enterprise single sign-on, and we are not working toward them right now. If your company needs those, Friday is not the right fit yet. See what Friday keeps and never gets on the security page. Are my own Claude Code, Codex and Cursor metered? No. They run on your computer under your own plans, on every Friday plan, and Friday never meters them. Does my code leave my computer? Your coding agents run on your computer under your own login, and Friday does not copy your repository to run them. Friday stores the session record (messages, commands and their output, and the changes made) so your team can replay it. The security page lists everything it keeps. How much frontier use does Team Pro include? Team Pro includes 550 credits of frontier models for Friday’s own agent in every 5-hour window, starting from your first message. Bigger models use credits faster, and the meter in the app shows what’s left and when it refills. If you run out, Friday’s own agent keeps going on its default model, which is unlimited on both Team plans, and your own coding agents are never counted. ## Run your next task in Friday. Free on your own, in your browser or on Windows and Mac. A new team gets Team Basic free for 7 days. No card required. [Try for free](/signup)[See pricing →](/pricing) Not at your computer? We’ll email you the link. One email, nothing else. No newsletter. --- Source: https://www.tryfridaywork.com/terms # Terms of Service Last updated 15 October 2026 ## 1\. Who we are Friday Work ("Friday", "we", "us") runs shared sessions where people and AI agents work on the same task together. These terms are the agreement between you and Friday Work for the web app at app.tryfridaywork.com, the Friday desktop app, the Friday runner you install on your own machine, and this website. Friday Work is not yet incorporated. Once it is, the company name, registered address and governing law will be named here, and you will be told before that change takes effect. By creating an account or using Friday you accept these terms. If you accept them for an organization, you confirm you may bind that organization, and "you" then means both you and it. ## 2\. What Friday does A session holds one or more agents. Each agent has one driver at a time: the person whose instructions it follows. Everyone else watches every step live, and may suggest a redirect, request control, or take over where the session allows it. The owner can always pause or cancel a run, even while someone else is driving. Everything that happens is appended to a session log: who instructed what, what the agent did, what each tool returned, and when. That log is what the room sees, what a latecomer catches up from, and what the audit trail is built on. Depending on the session, agents can search the web, write documents, slide decks and spreadsheets, call tools you connect over MCP, and — when you link your own machine — read and change files in a folder you choose and run commands in it. **Beta and changing features.** Friday is young and changes often. Features marked early or experimental may change or disappear without notice. We will not remove a feature your paid plan describes without telling you first. ## 3\. Accounts You sign in with Google, with GitHub, or with an email address and password. Keep your credentials and your session links safe, and tell us promptly at security@tryfridaywork.com if you believe an account has been used without permission. You are responsible for everything done through your account, and for what the people you invite do in your workspace. You must be old enough to enter a contract where you live, and at least 13 years old. Friday is built for work, not for children. One person, one account. Do not share a single login between several people — invite them to the workspace instead. ## 4\. Workspaces, sessions and invitations Work lives in a workspace. Its owner decides who is a member, which plan it runs on, and which connections its agents may use, and can see and manage everything in it. **Anyone who can open a session can read all of it**, including what happened before they joined. A session is a room with a transcript, not a private message. **Invitation links are keys.** An editor link lets anyone holding it who can sign in steer agents in that session; a viewer link lets anyone holding it read the session without an account. Share them as carefully as the work itself. If your workspace belongs to an employer or a client, that organization controls its content and may add or remove members and read what is in it. ## 5\. Your content and ours You keep ownership of what you and your team put into Friday: instructions, uploaded files, documents agents build with you, and the code in folders you attach. As between you and us, you also own what agents produce in your sessions, so far as such output can be owned. You grant us a worldwide, non-exclusive, royalty-free license to host, copy, transmit and display that content, and to send the parts an agent needs to the model and tool providers named in our Privacy Policy. The license exists only to run the service for you and ends when the content is deleted. You confirm you have the rights to the content you put into Friday, including personal data about other people, and that using it in Friday breaks no law or agreement you are under. We own Friday itself: the software, the interface, our name and our marks. Nothing here transfers that to you. **Feedback.** If you send us ideas for improving Friday we may use them freely and without owing you anything, so please do not send anything confidential as feedback. ## 6\. Acceptable use Do not use Friday to break the law, to infringe anyone’s rights, or to build or operate anything designed to harm people. In particular, do not use it to create malware, to attack systems you do not own or have written permission to test, to harass anyone, or to produce sexual content involving children. Do not work around plan limits, probe or disrupt our infrastructure, resell Friday without our written agreement, or use it to build a competing model from data you do not own. Do not upload content you have no right to upload, and do not connect a tool you are not authorized to connect. We may suspend an account that breaks these rules, or where suspension protects other customers, our infrastructure, or compliance with the law. Where we reasonably can, we will tell you first and give you a chance to put it right. ## 7\. Agents, your machine and your own CLI **Agent output can be wrong.** Agents produce text, documents and code that may be inaccurate, out of date or unsuitable for your purpose. Nothing Friday produces is professional, legal, financial or medical advice. Review anything that matters, and read code before you ship it. **You choose what an agent may do on your machine.** When you link a machine you pick the mode: plan only, edit files, or edit files and run commands. Friday will not exceed the mode you chose — but within it, an agent can change or delete files and run commands with real effects. Point it at a folder under version control, keep your own backups, and do not link a machine holding anything you cannot afford to lose. **Your own CLI stays yours.** A coding session can run on a CLI you already have, such as Claude Code, Cursor, or Codex, on your machine and under your own subscription. We never hold your credentials for it, we never meter it, and that provider’s terms govern your use of it. **Connected tools act with the access you grant.** An agent can use a connected tool within the permissions you approved, including tools that change data. Only tools a server marks read-only can be stopped mid-call; anything else finishes, and the log records what happened. ## 8\. Third-party services Friday sends prompts to AI models through OpenRouter, which routes them to providers including Anthropic, OpenAI, Google and DeepSeek. Agent web searches go out the same way. We are not responsible for third-party services, their availability, or what they do with data you direct to them. Your use of them is governed by their own terms. ## 9\. Plans, usage and payment Solo, Team Basic and Team Pro include what the pricing page describes on the day you subscribe. Solo is free. It runs the coding tools you connect, such as Claude Code, Codex and Cursor, on your own accounts with their providers. On Solo, Friday’s own agent runs only on a model you bring, with your own key or on your own computer, and Friday does not meter it. Team Basic and Team Pro are priced per seat and include Friday’s own agent: on Friday’s default model on Team Basic, with frontier models added on Team Pro. Frontier-model use on Team Pro is metered in tokens over 5-hour windows. Friday’s default model is not rationed on either Team plan, and a coding tool you run yourself is never metered on any plan. Running out of Team Pro’s frontier credits does not end your session: agents keep working on Friday’s default model, or on a coding tool you connect, until the window ends and the allowance starts over. A team may get Team Basic free for 7 days, without a payment method; the 7 days start when a teammate who can edit first joins or you invite one to edit, or when you first run Friday’s agent, and each account gets one trial. When the trial ends, the workspace moves to Solo unless you choose a paid plan. Paid plans are billed in advance, per seat, monthly or yearly, and renew automatically until you cancel. Cancel at any time and your plan runs to the end of the period you have paid for. We do not refund part-periods unless the law where you live requires it. We may change prices with at least 30 days’ notice before a renewal. Taxes are added where they apply. If a payment fails we may move the workspace to Solo after telling you. ## 10\. Availability and support We work to keep Friday up, but we do not promise a particular level of uptime and we may take it down for maintenance. Planned maintenance is announced in the app where we can. Support is by email, and we answer as quickly as we reasonably can. ## 11\. Ending the agreement You may stop using Friday and close your account at any time. We may end this agreement with 30 days’ notice, or immediately if you seriously break these terms. When an account ends we stop billing it, and the retention rules in the Privacy Policy decide how long anything is kept. Export what you need first: sessions export as documents, and as a readable audit of who instructed what. If we close your account without cause, we refund the unused part of what you paid. ## 12\. Warranties and liability Friday is provided “as is” and “as available”. To the fullest extent the law allows, we exclude implied warranties including merchantability, fitness for a particular purpose and non-infringement, and we do not warrant that Friday will be uninterrupted or error-free, or that agent output will be accurate. To the fullest extent the law allows, neither party is liable for indirect, incidental, special or consequential loss, for lost profits, revenue or goodwill, or for lost or corrupted data. Our total liability for all claims in any twelve-month period is limited to the greater of what you paid us in that period and one hundred US dollars. Nothing here limits liability that cannot lawfully be limited, including for death or personal injury caused by negligence, or for fraud. You will cover us against claims by others arising from content you put into Friday, from your breach of these terms, or from your use of a connected tool or machine. ## 13\. General **Changes.** We may update these terms. If a change materially affects you we will tell you at least 30 days before it takes effect, by email or in the app, and continuing to use Friday after that date means you accept it. Earlier versions are available on request. **Notices.** We contact you at the email address on your account. You reach us at legal@tryfridaywork.com. **Transfer.** You may not transfer this agreement without our consent. We may transfer it to a company that acquires our business, on the same terms. **Whole agreement.** These terms and the Privacy Policy are the entire agreement between us about Friday. If part of them is unenforceable, the rest still applies. Not enforcing a right once does not waive it. **Governing law.** Once Friday Work is incorporated, the governing law and the courts that hear disputes will be named here. Until then, nothing in these terms takes away rights you have under the law where you live. --- Source: https://www.tryfridaywork.com/privacy # Privacy Policy Last updated 15 October 2026 ## 1\. The short version We collect what we need to run shared agent sessions: who you are, what your team asked agents to do, what those agents did, and how much model usage a workspace has spent. **We do not sell personal data. We do not use your session content to train AI models, and we ask our model providers not to either. We show no adverts and do not use advertising or cross-site tracking cookies.** Session content goes to AI model providers so agents can answer, and to any tool you connect so agents can use it. Everything else stays with us and the infrastructure we run on. Questions, or a request about your data: privacy@tryfridaywork.com. ## 2\. Who is responsible for your data For your account and for this website, Friday Work is the controller of your personal data. For what your team puts into a workspace, the organization that owns the workspace is the controller and Friday Work is its processor, acting on that organization’s instructions. If your employer gave you the workspace, ask them first about content in it. ## 3\. What we collect **Account details.** Your name, email address and profile picture from Google or GitHub, or the name, email address and password you give us directly. Passwords are stored only as a salted scrypt hash, never in a form anyone can read. **Session content.** Every instruction, suggestion, message, question and answer in a session; the documents, decks, sheets and code agents write; files you upload; and what each tool call returned. It is kept as an append-only log, because showing who asked for what is the point of the product. **Workspace details.** Workspaces, members and roles, invitations, projects, and the custom agents and skills you create. **Connections.** For each MCP connection: the name and URL you gave it, and the access token from your sign-in with that provider. **A download link you ask for.** If you ask the website to email you the download link, we keep that email address to send the link once and to avoid sending it twice. We do not use it for anything else or add it to a mailing list unless you ask us to. Email privacy@tryfridaywork.com to have it deleted. **Usage.** The model tokens a workspace uses in each 5-hour window, so plan limits apply, and who is present in a session. **Technical data.** Ordinary server logs, including IP address, browser and timestamps, kept for security and debugging. **If you link your own machine.** Which coding CLIs it has, the folders and repositories you choose to expose, the changes an agent makes there, and the output of commands it runs. We never receive your credentials for that CLI. **What we do not collect.** We do not take payment card numbers ourselves — a payment processor handles those. We do not track you across other websites, and we do not buy data about you. ## 4\. Cookies and local storage Friday sets two cookies that are strictly necessary: a signed session cookie proving you are signed in, and one recording which workspace you are viewing. Both are HttpOnly and both expire. Neither is used for advertising. **One more, only if you arrive by a Friday link.** If you open a shared session link, or reach the app from a link that carries a source tag, the app sets a 30-day cookie called fw\_ref. It holds only a short label for how you arrived (for example watch or replay) and the session that link pointed to. It contains no identifier for you, and it is used only to count how people reach Friday and where a new account came from. It is not used for advertising. **Sign-in flow cookies.** While you sign in with, or connect, a provider such as Google or GitHub, the app sets short-lived cookies (friday\_connect, friday\_oauth, friday\_separate) that tie each step to the browser that started it. They are strictly necessary and carry no tracking. **Analytics inside the app.** When you use the Friday app, its server sends usage events to PostHog, tied to your account id: event names, counts, ids and agent names. Prompts, file paths, commands and code are never sent. You can turn this off under Share usage analytics in Settings. **Analytics, and your choice.** On our public website we use PostHog to see how the site is used: which pages are viewed, which sign-up, log-in, pricing and download links are clicked, how far down a page people scroll, how long it takes someone to do anything, which questions on the home page they open, how quickly pages load, and the campaign tags (utm\_source, utm\_medium and utm\_campaign) in the link you arrived by. Until you choose, and if you choose necessary cookies only, every visit is counted anonymously: nothing is stored on your device, no identifier is created, and one visit cannot be told from the next. We may also ask one optional question about what brought you here; your answer, and a few words if you choose to type them, are sent to us with the visit. If you allow all cookies, PostHog stores a random identifier in a first-party cookie and in your browser’s local storage so one visit can be told from the next, and we record short replays with everything you type hidden. We also record the sign-in and sign-up pages this way, with what you type hidden, to see where people get stuck; we never record the account page or the pages for verifying, resetting or connecting. Replays are deleted after 30 days. **When you are signed in,** your account page sends PostHog your account ID, email address, plan and how you signed up, so we can see which visits led to an account and connect them with how you use the app. If your browser sends "Do Not Track", nothing is collected. Data goes through our own domain to PostHog in the United States. To change your answer, clear this site’s data in your browser and you will be asked again. **This tab only.** The home page remembers which of our public pages you have already opened in the same browser tab, so its opening can match (for example, if you looked at pricing, it puts "Try for free" first). This is kept in the tab's own session storage on your device, is never sent to us, and disappears when you close the tab. Your browser also keeps your own sidebar — recent sessions, projects and your private labels for them — in local storage on your device. Clearing your browser data removes it. ## 5\. How we use it, and why we may To run sessions and show them to the people you share them with; to send prompts to model providers and tools so agents can work; to apply plan limits; to keep the audit trail the product promises; to answer support requests; to keep accounts secure and detect abuse; and to bill paid plans. **Legal bases (UK/EU).** Performing our contract with you, to run and bill the service. Our legitimate interests, for security, abuse prevention and improving Friday, balanced against your rights. Legal obligation, where we must keep records. Consent, where we ask for it — such as product emails you can unsubscribe from at any time. We look at aggregated, non-identifying usage figures to understand how Friday is used. We do not read your session content except when you ask us to for support, when we must to investigate abuse, or where the law requires it. ## 6\. Who we share it with **People you share with.** Everyone in a session can read its whole log, and everyone in a workspace can use agents and skills shared at workspace level. **OpenRouter**, which routes prompts to model providers including **Anthropic, OpenAI, Google and DeepSeek**. Agent web searches go the same way. **Tool providers you connect** over MCP — Google Workspace, Slack, GitHub, HubSpot and others — receive only what an agent’s tool call sends them, using the access you granted. **Infrastructure:** our hosting provider, our database, and object storage for uploaded files where configured. **Payment processing** for paid plans, which takes your billing details directly. **Authorities**, where the law genuinely requires it, and a buyer of the business if Friday is ever sold — in which case you will be told before your data moves. Each of these works under a contract limiting them to what we ask of them. We do not share personal data with advertisers. ## 7\. How long we keep it The app shows the session history your plan includes: at least 30 days on Solo, and at least 365 days on Team Basic and Team Pro. Older sessions are not yet deleted automatically (see below). Detailed usage is kept for 100 days, and daily totals (counts only) for about a year; both are deleted with the account. Account details are kept while the account exists, then deleted. Backups are kept briefly and overwritten in turn, so deleted content can persist there for a short time before it is gone for good. **Deleting your account.** You can delete it yourself under Delete account on your account page. That cancels any plan, deletes the workspaces you own alone with their sessions, removes you from other workspaces with your personal agents and memory in them, and deletes the account itself. What you shared with a team, such as your messages in its sessions, stays with that team. Website analytics already recorded are deleted when you ask us. If you are the only owner of a workspace other people are in, make someone else an owner or remove them first. You can also email privacy@tryfridaywork.com and we will do it for you, at no charge. Automatic deletion of expired sessions is still being built. ## 8\. How we protect it Traffic is encrypted in transit. Sign-in cookies are signed, HttpOnly and time-limited. Passwords are salted and hashed with scrypt, and repeated failed sign-ins on an address are rate-limited. Access to a session is checked on every action, and the control token that decides who may steer an agent is enforced by a single atomic database operation, so permission checks cannot be raced. Access tokens for tools you connect are stored inside your workspace in our database. They are not yet encrypted at rest; that work is planned, and we would rather say so than imply otherwise. If that matters to your organization, wait for it before connecting sensitive systems. No service is perfectly secure. If you find a vulnerability, report it to security@tryfridaywork.com before disclosing it publicly and we will work with you on a fix. ## 9\. Your rights Depending on where you live, you may ask for a copy of your personal data, correct it, delete it, restrict or object to how we use it, or receive it in a portable form. You may also withdraw consent where we relied on it. Sessions can be exported at any time as documents, and as a readable audit of who instructed what, so a copy of the work is always in your hands. Email privacy@tryfridaywork.com and we will answer within 30 days. We will not charge you or treat you differently for asking. If our answer does not satisfy you, you can complain to your local data protection authority. Where your workspace belongs to an organization, we pass your request to them and help them answer it. ## 10\. International transfers Friday and the providers above process data in the United States and other countries. Where personal data leaves the UK or the EEA we rely on standard contractual clauses or another lawful transfer mechanism, alongside the protections described here. ## 11\. Children Friday is for work and is not directed at children. We do not knowingly collect personal data from anyone under 13. If you believe a child has given us data, email privacy@tryfridaywork.com and we will delete it. ## 12\. Changes to this policy We post any change on this page and update the date at the top. If a change materially affects how we use your personal data, we will tell you in the app or by email before it takes effect. --- Source: https://www.tryfridaywork.com/about # Agents are better with a team around them. ## Why we built it Coding agents are very good when one person is steering one of them. The moment a team is involved, it falls apart. One teammate has Claude Code open, another is in Codex, a third is in Cursor. Nobody can see what anyone else's agent is doing, nobody can step in halfway through a run, and when an agent gets stuck, its context is stuck with it. Friday Work is a shared workspace where the people and the agents work in the same session. You share a session, teammates join and watch live, and anyone can take the wheel. You can run your own coding tools beside Friday's agents, each in its own worktree, hand work from one to another, and carry on from a Claude Code or Codex session you already had. ## What we believe ### See every step An agent that works out of sight is an agent you have to trust blindly. In Friday, everyone in the session watches every step as it happens, and the whole run is kept as a record you can scroll back through. ### Someone is always driving Each agent has one driver at a time, so there is never a question of whose instructions it is following. Anyone can ask for the wheel, and the owner can always stop a run. ### Your tools stay yours Claude Code, Codex and Cursor run on your own machine, under your own accounts. We never hold their credentials and we never meter them. Friday is where they meet your team, not a replacement for them. ## Who is behind it Friday Work is built by Arun Dass, its founder, who started it because he kept running into this problem himself. It is a small, young product and it changes quickly. We read every message we get, and a lot of what is in the app today came from someone telling us what was in their way. ## Try it, and tell us what you think Friday is free to start, in your browser or on Mac and Windows. If something is confusing, missing or broken, we want to hear it. [Try for Free](/signup)[Get in touch →](/contact) --- Source: https://www.tryfridaywork.com/contact # Talk to us. One form for everything. We read every message and reply to the email you give us. What is it about? QuestionSupportFeedbackSalesSecurityPrivacy and your dataLegalPress Your nameEmail Message Rather use email? Write to [hello@tryfridaywork.com](mailto:hello@tryfridaywork.com). --- Source: https://www.tryfridaywork.com/security # What stays with you, what we keep, and what is not done yet. ## What Friday collects, in short ### What we keep Your account Your name, email and sign-in. Passwords are stored only as a salted hash. What happens in a session The messages, each agent’s steps, the commands it ran and what they printed, and the changes it made, so your team can watch live and look back. Usage for your plan How much of Friday’s own agent a workspace has used, so plan limits work, and who is in a session. Usage events The app sends usage events to PostHog, tied to your account id: event names, counts, ids and agent names, never prompts, paths or code. Turn it off under Share usage analytics. Server logs Ordinary logs (IP address, browser, time) for security and debugging. ### What we never get - Your repository: your coding agents work on it on your computer, and Friday does not copy it. - The sign-ins of your coding tools: Claude Code, Codex and Cursor run under your own accounts. - Card numbers: payments go through Stripe. - Ad trackers or cross-site tracking. The website counts visits anonymously, and records a visit only if you say yes. Beyond sign-in and workspace, the app sets fw\_ref when you arrive by a shared link (a 30-day label for how you arrived, with no identifier for you) and three short-lived cookies, friday\_connect, friday\_oauth and friday\_separate, that last only while you sign in with or connect a provider such as Google or GitHub. ## What stays on your machine Claude Code, Codex and Cursor run on your own computer, under your own accounts. Friday never holds their sign-in keys, and it never meters them. Your files are read and changed by those tools on your machine, in the folder you chose. Friday does not copy your repository to run them. ## What Friday stores What an agent does in a session is stored so the team can watch it live and wind it back later: the messages, the agent’s replies, the commands it ran and what they printed, and the changes it made. If an agent prints a secret, that line is in the record, so keep secrets out of the folders you point an agent at. If you attach a folder to a session that has no machine connected, that working copy is stored on Friday’s server for the session. Your account details are stored, and card payments are handled by Stripe, so a card number never reaches Friday. ## Before you share a session Everyone in a session can read its whole record, including anything an agent printed. Keep secrets out of the folders agents work in, invite only people who should see the work, and turn a link off when you are done with it. ## Who can do what in a session Everyone in a session is an owner, an editor or a viewer. Viewers watch, and a viewer link opens without an account, read-only. Editors sign in, and can suggest and, when they hold the wheel, steer. Each agent has one driver at a time, and an owner can always stop a run. Invite links can be turned off or replaced at any moment, and a person you remove is out at once: they are disconnected, they cannot come back with the link, and they lose the control of any agent they were driving. Access is checked on every action, not only when someone joins. ## What an agent may do Plan reads and proposes, asks before it runs a command, and changes nothing. Manual asks before any step that changes something. Auto edits and runs commands without asking, for work you trust. The person whose machine it is changes this, and everyone in the session can see which one is on. A request to run something goes to the person driving that agent, and the whole session sees it. ## How it is protected Traffic is encrypted in transit. Sign-in cookies are signed, HttpOnly and time-limited. Passwords are salted and hashed with scrypt, and repeated failed sign-ins are rate-limited. The control token that decides who may steer an agent is enforced by a single atomic database operation, so permission checks cannot be raced. ## What is not finished Access tokens for tools you connect, such as GitHub or Slack, are stored in your workspace in our database and are not yet encrypted at rest. That work is planned. If it matters to your organization, wait for it before connecting sensitive systems. You can delete your account yourself under Delete account. Automatic deletion of expired sessions is still being built; until it ships, email privacy@tryfridaywork.com and we will delete your sessions by hand, at no charge. Friday is built for individual developers and small teams, not for enterprises. We have not had an independent security audit, we do not hold SOC 2 or ISO 27001, and Friday has no enterprise single sign-on (SAML or SCIM). We are not working toward them right now, and we will say so here if that changes. If your company needs a certified vendor, Friday is not the right fit yet. ## How we look at the website The public website counts every visit anonymously, with nothing stored on your device. There are no advertising or cross-site trackers. What is collected, and the two optional questions we may ask, are set out in the privacy policy. ## Found something, or need an answer? Report a vulnerability to security@tryfridaywork.com before disclosing it publicly and we will work with you on a fix. For anything about your data, write to privacy@tryfridaywork.com. [Get in touch](/contact)[Read the privacy policy →](/privacy) --- Source: https://www.tryfridaywork.com/changelog # What’s new in Friday. The biggest updates, newest first. 1. 15 October 2026 ## A calmer coding session, with named agents - Coding sessions open in a calmer Focus layout, and agents have names. - Home, the Agent Board and the Usage hub are redesigned, with an activity graph. - If a message looks like it contains a key, Friday tells you before sending. - Desktop 0.1.56 for Windows, with security improvements. 2. 13 October 2026 ## Handoffs that carry the whole record, and a faster, steadier session - Handoffs give the next coding tool everything the first one already knows. - Another coding tool you pick joins on its own lane on the same code. - End-of-answer questions get an answer card and a Waiting on you toast. - Answers read like a docs page, with tables and highlighted code. 3. 12 October 2026 ## Codex streams, hubs, and every agent at work in one place - Codex streams its answers live, and asks about each change in Manual. - Hubs plan your idea as tasks, and several agents build it at once. - Smart routing grades each run on evidence and learns what works per repo. - The Agents hub shows a card per agent with its plan and usage. 4. 11 October 2026 ## Smart routing: the right agent and model for every task - Friday picks the model most likely to get each task right for less. - Auto in Claude Code and Codex picks the model for each message. - The agent picker suggests a tool and model as you type. - At a limit, Friday offers the best other coding tool on your computer. 5. 11 October 2026 ## GitHub Copilot and ten more coding tools, your own models, and a project brain - GitHub Copilot, Gemini CLI, Cline and more join Claude Code, Codex and Cursor. - Coding tools stay running between messages, so they answer in seconds. - Friday’s agent can use your own API key or a local model. - A project brain briefs the next agent with what earlier agents learned. 6. 10 October 2026 ## Solo and Team, Code Review, and your folder left alone - Solo runs your own coding tools for free, and Team brings your team. - Code Review for pull requests runs right on your computer. - Agents work on a copy of uncommitted folders until you press Apply. - Undo puts back only what that agent run changed. 7. 8 October 2026 ## Any coding tool can carry on from any other - Start in Claude Code and carry on in Codex, Cursor or Friday Code. - One box sets the agent, model, branch and Plan, Manual or Auto mode. - Teammates can bring their own agent without having the session’s folder. - Owners can remove people from a session. 8. 2 October 2026 ## Plan, Manual and Auto - Plan, Manual and Auto set how much an agent may do alone. - Hand off work between agents and fork a session or message. - See your coding tools’ usage limits beside the session. - The agent builder writes onto the canvas as it thinks. 9. 1 October 2026 ## Teammates bring their own coding agent - Each teammate brings their own Claude Code, Codex or Cursor. - A built-in browser previews the site your agent is running. - When a model fails, Friday retries on the next one. - Cloning a GitHub repository is more reliable and explains failures. ## Get every update the day it ships. No payment required. Works in your browser. [Try for free](/signup)[Every release, with smaller fixes, on GitHub →](https://github.com/splashxmoon/friday-releases/releases)