Installation

Crew ships as a Python package that bundles the React dashboard. There are four ways to install it — pick whichever fits how you'll run it. All builds are driven by plain pip + npm/Vite + pytest. No proprietary tooling.

Crew drives an LLM through kiro-cli over the Agent Client Protocol. It is the only provider (agent.provider = acp). The first desktop or dashboard launch can install Kiro CLI on the Gateway host and guide device-code sign-in before chat opens. kirocrew doctor reports whether the binary is found and logged in.

Semantic memory needs no setup. Embeddings run in-process and the Gateway downloads its embedding model (~610 MB) in the background on first start, verifies it, and stores it under ~/.kiro/crew/models. Until the model lands, memory search falls back to keyword matching and picks up embeddings automatically without a restart. Set KIROCREW_EMBED_MODEL_URL to point at a mirror for airgapped installs.

Crew supports Windows natively. Use CPython 3.12 + a venv + pip install -e . tzdata, and launch with python -m kiro_crew gateway.

py -3.12 -m venv .venv
.venv\Scripts\activate
pip install -e . tzdata
python -m kiro_crew gateway

All POSIX-only process, signal, file-lock, and metrics calls are routed through kiro_crew.platform_compat. The OS-level filesystem sandbox (Linux namespaces / macOS Seatbelt) is not available on Windows; every other feature works, and Computer Use drives native Windows applications through UI Automation. The interactive web-terminal and the SSO-login PTY degrade to a clear "not supported on Windows" response rather than crashing.

For the one-line wheel, source build, and Docker installs, set up and verify after installing:

kirocrew setup            # interactive wizard: data dir, agent, credentials
kirocrew doctor           # verify everything is wired up
kirocrew gateway          # start the server → http://localhost:5476

kirocrew doctor reports the status of the kiro-cli binary, its authentication, Ollama/embeddings, Slack tokens, config validity, and MCP servers.

All persistent state lives under ~/.kiro/crew/ (override with KIROCREW_HOME):

~/.kiro/crew/
├── config.json          # user configuration
├── .env                 # Slack tokens, owner ID, other credentials
├── workspace/
│   ├── memory/          # preferences.md, projects.md, history/
│   ├── lessons.jsonl    # learned corrections
│   └── knowledge/       # ingested documents (FTS5 + vectors)
├── conversations/       # JSONL session logs
├── crons.json           # scheduled jobs
├── audit.log            # bash command audit trail
└── agents/              # generated kiro-cli agent configs

Both KIROCREW_HOME and KIROCREW_PORT are validated at CLI entry, not stored as config keys. Pass --port on the CLI to override the env var per-invocation.

Install with pip install "kirocrew[<extra>]":

Piper TTS is the default text-to-speech provider and runs locally — no extra install needed.