Crew is an open-source personal AI agent that runs locally or remotely on your hardware. It is persistent, self-learning, and self-evolving. You can work with it from the Mac app, web dashboard, and CLI, or additional channels such as Slack, Discord, Telegram, Teams, Webex, WeCom, and WeChat (Weixin). Multi-step tasks run unattended, recurring jobs run on your schedule, and heartbeats monitor systems until something needs attention.
Download Crew for macOS (signed .dmg) or Linux (.AppImage). The app starts a bundled Gateway when no local Gateway is already running. It can also connect to a remote Gateway over an SSH tunnel.
Install the prebuilt, SHA-256-verified wheel from the release CDN:
curl -fsSL https://download.crew.kiro.dev/cli.sh | sh
The installer installs through pipx when available or a managed venv at ~/.kiro/crew/venv, and records the channel in ~/.kiro/crew/channel.
After installing, configure and start the Gateway:
kirocrew setup # interactive wizard: data dir, agent, credentials kirocrew doctor # verify everything is wired up kirocrew gateway # start the server → http://localhost:5476
Open http://localhost:5476 in your browser. The first launch guides you through installing kiro-cli and completing sign-in. That's it — you're ready to chat.
Semantic memory needs no setup. Embeddings run in-process and the Gateway downloads its model in the background on first start. Until it lands, memory search falls back to keyword matching and picks up embeddings automatically without a restart.
If you prefer to build from source, run in Docker, or install on Windows, see the full Installation guide.
Once the dashboard is open (either from the Mac app or at http://localhost:5476), here are a handful of workflows to explore.
Type in the chat panel and press Enter. Each tab is an independent session with full tool access — read/write files, run commands, browse the web, or call any configured MCP tool. Sharpen a prompt with the prompt optimizer (Cmd+Shift+Enter).
Open the Projects panel, describe what you want, refine it into a spec, and hit Run. The Task Runner decomposes the spec into steps, executes them, runs tests, retries on failure, and checkpoints progress.
Open the Schedule panel and add a job with natural language — or just ask in chat: "Every weekday at 9, summarize my open work." Jobs run in their own sessions. Results land in your dashboard notifications, Slack DM, or Telegram chat.
Tell Crew something it should always remember in any chat session: "No, always run the frontend checks before calling a change done." It saves the correction as a lesson and applies it in future sessions.
Ask in chat: "Research these three options in parallel and recommend one." Crew spawns isolated subagents and synthesizes the results back into your conversation.
Ask Crew to watch a system — in any chat session. Heartbeats check in on a schedule and alert when something needs attention. No manual polling loop required. Set one up from chat or via a cron job.
Persistent and always available. Sessions, memory, schedules, and task checkpoints persist beyond one chat and across Gateway restarts. Scheduled, proactive, and reactive work continues without someone at the terminal.
Self-learning. Corrections and task failures become durable lessons that change later behavior. Preferences and project context carry across sessions.
Self-evolving. Repeated patterns can become reusable skills. Memory, lessons, and skills remain visible and editable, so your Crew becomes more tailored to you and your work over time.
Runs where you choose. Keep the runtime and its state on your Mac, in a local container, or on a remote machine you control.
One Gateway, many ways to work. The Mac app and web dashboard let you work directly. The Gateway connects that same runtime to Slack, Discord, Telegram, Teams, Webex, WeCom, WeChat (Weixin), and the CLI when you want to continue from another surface.
Quick start