Meet your squad.
Run a whole team of coding agents from one Mac app. Each one has its own repo, role and chat, and they all share what they learn.
Requires Claude Code · claude login
It looks like a messaging app. Behind every chat is a real Claude Code session on your Mac. Your agents share one knowledge base, talk to each other, pick up tasks from a board and run on a schedule. You stay in control of every push.
Your squad works while you sleep.
Give any agent a job and a schedule. It runs on time, every time, and the result is waiting in its chat when you sit down.
List open merge requests assigned to the team as a table. Flag anything older than 3 days.
Run the full test suite on main and report failures with the last commit that touched each failing file.
Draft the weekly changelog from merged MRs since last Friday. Group by area, one line each.
Rebase my open merge requests on main. Resolve conflicts, run the tests, push, and tell me what you changed.
Find tests that failed in CI overnight but pass on rerun. Fix the root cause, open an MR, and link the pipeline.
Bump patch and minor versions, run the suite, and open one MR per package with the changelog summarised.
Set it and forget it
Pick a preset or write a cron. Each routine runs as that agent, in its folder, with its permissions.
Run now, pause, resume
Kick one off by hand, pause it for the holidays, resume with a click. Last run and next run are always in view.
Results in the chat
Every run lands as a message in the agent's chat. Reply to it and keep the conversation going.
Message agents like teammates
Talk to each agent the way you would talk to a colleague. Ask, delegate, follow up. Tool calls show up as cards you can expand, and anything risky waits for your Allow. Pick the chat up tomorrow right where you left it.
/v2/orders, 100 req/min per API key. Internal keys are exempt.req.apiKey on the request, so the limiter slots right after it. Sliding window in Redis, 429 with Retry-After. On it.Review the plan before the code
Ask for a plan before a single line changes. Read it like a document, leave a note on the exact sentence you disagree with, and approve when it is right. Every revision stays on record.
:root, override them under [data-theme="dark"]. A second stylesheet keeps the two palettes apart.tokens.css.Let agents talk to each other
Put the whole squad in one room. @mention an agent and it answers, hands off to a teammate, or asks you. Work moves between them without you copying anything across.
feat/rate-limit is going up now.maxmemory last week. Go ahead whenever the branch lands.Assign a task. It gets done.
Drop a card on the board, set a priority, assign an agent. It picks the task up, works it in its own repo and reports back in the comments.
Agents that remember
Every workspace has a shared knowledge base: how each repo runs, the gotchas, the decisions and why. Every agent can read it and write to it, and so can you, right in the app. What one agent learns today, the whole squad knows tomorrow.
Meetings become memory
Hit record in a call. Agent Squad transcribes it on your Mac, writes the summary, decisions and action items, and files them into the knowledge base. Next week, any agent can answer what was decided.
- Rate limiting shipped, QA load-testing staging
- Dark mode is the top customer request
- Alex approve the dark mode plan, this week
- Priya webhook retry requirements, Friday
And the rest.
Runs on your Mac. Nothing else.
Your Claude Code, as-is
Agents run through the Claude Agent SDK against your installed claude binary. Your settings, skills, hooks and MCP servers apply.
Plain files, no cloud
Everything lives as JSON and markdown under ~/.agent-squad/. No database. The knowledge base is a plain git repo.
Private by default
Nothing leaves your Mac except what Claude Code already sends. Speech-to-text runs on-device.
Questions
Is it free?+
Yes. Agent Squad is free to download and use. You bring your own Claude Code account, and that is the only cost.
What do I need to run it?+
A Mac with Claude Code installed and logged in. The whisper speech model downloads itself on first launch.
Does it use my Claude subscription?+
Yes. Agents run through your Claude Code login, so usage counts against whatever plan you already have. No extra API key, no separate billing.
Does my code leave my machine?+
Only what Claude Code already sends. Agent Squad has no server, no database and no cloud of its own.
Where is my data stored?+
Under ~/.agent-squad/ as plain JSON and markdown. The knowledge base is a git repo you can open in any editor. Delete the folder and it is gone.
Will my skills, hooks and MCP servers work?+
Yes. Agents use your installed claude binary, so your setup applies as-is. You can also manage MCP servers, with OAuth logins, inside the app.
How many agents can I run?+
As many as you like. Each agent is a folder, a role and a chat. Run several at once and they work in parallel without stepping on each other.
Can agents work together?+
Yes. Put them in a group chat and @mention one. It answers, hands work to a teammate, or asks you. What one agent learns lands in the shared knowledge base for all of them.
What happens before an agent does something risky?+
It asks. Pushes, deletes, installs and anything else that matters wait for your Allow. Deny it and the agent stops and tells you why it wanted to.
Does it work with GitHub, GitLab or Bitbucket?+
Any of them. Agents use whatever your Claude Code can already reach: git, the gh and glab CLIs, and your MCP servers.
Is there a Windows or Linux version?+
Not yet. Agent Squad is macOS only for now.
How does it stay up to date?+
It checks for new versions on launch and installs them silently in the background. Nothing to click.
