Skip to main content
Crewship is a self-hosted runtime for AI coding agents. This page gets you from zero to a running agent in two passes: a 60-second smoke test that proves the runtime is alive on your hardware, then a 5-minute full setup that creates your first crew.
v0.1 beta — use the Claude Code adapter. Crewship ships scaffolds for Codex, Gemini, OpenCode, Cursor, and Factory Droid, but only Claude Code has production-tested streaming, MCP support, and prompt scrubbers in this release. Pick Claude Code in the agent picker for the supported path; the other adapters are marked Beta in-product and have known gaps (Cursor MCP is upstream-broken in headless mode; the Codex live-event stream parser is still a stub). See the README’s “Beta status & limitations” section for the full matrix.

60-Second Smoke Test

If you just want to see the dashboard come up on your machine, this is enough:
# 1. Install (binary, single command)
curl -fsSL https://raw.githubusercontent.com/crewship-ai/crewship/main/scripts/install.sh | bash

# 2. Start the runtime — embedded UI + SQLite + Go server, no Node.js, no external DB
crewship start

# 3. Open the dashboard
open http://localhost:8080
That’s it. The runtime is now serving on your hardware. No agents have been created yet — that’s the next pass.

Prerequisites for the Full Setup

  • Docker (or Apple Containers on macOS 26+) — one Linux container per crew, shared by its agents
  • An LLM API key (Anthropic, OpenAI, or Google) — agents call out to whichever provider you wire up
  • Go 1.26+ and Node.js 22+ + pnpm — only if you build from source instead of using the binary

Installation

brew install crewship-ai/tap/crewship

crewship version
For pinned versions, custom install directories, and air-gapped install paths see the full Install guide.

Initial Setup

1

Start the server

crewship start
The server starts on http://localhost:8080 with an embedded UI and SQLite database.
2

Create the first admin user

Two first-class ways — pick the one that matches your deploy:
  1. Open http://localhost:8080/login auto-redirects to /bootstrap on a fresh instance.
  2. Fill in name + email + password (min 8 chars), click Continue to workspace setup.
Single-form bootstrap, no token to copy from anywhere.No rush by default. The bootstrap page stays open until you create the first admin — the empty user table is the gate, so take your time (the same first-run pattern as GitLab/Grafana). Once an admin exists, /bootstrap is closed for good.
Hardening for public instances (opt-in). If you expose a fresh instance to the internet before bootstrapping it, set CREWSHIP_BOOTSTRAP_WINDOW=5m (any Go duration) to arm a finite deploy-race window: /bootstrap accepts requests only for that interval after startup, then returns 410 Bootstrap window expired — restart the server to open a new one so a scanner can’t grab the admin account. Both the browser form and crewship init POST the same /api/v1/bootstrap endpoint and share the window; for unattended provisioning that lands minutes-to-hours after boot, restart the server immediately before the call so the window is fresh. Left unset (the default), bootstrap simply stays open until the first admin exists.
3

Log in via CLI

crewship login
Authenticates your CLI session against the server.
4

Select a workspace (multi-workspace setups)

crewship init creates your first workspace and marks it as the default — you can skip this step on a single-workspace install. If you belong to more than one workspace, pick which one the CLI targets for every subsequent command:
crewship workspace list
# SLUG         NAME          ID        ROLE
# acme *       Acme Corp     ws_abc…   OWNER
# side-proj    Side Project  ws_def…   ADMIN

crewship workspace use side-proj
# ✓ Default workspace set to: side-proj
The selection is written to ~/.crewship/cli-config.yaml and used by every subsequent CLI call. Override per-command with --workspace <slug> if needed.
5

Check system health

crewship doctor
Verifies Docker is running and all runtime dependencies are met. The sidecar binary is embedded in the crewship binary at build time, so no separate pre-build step is required.
6

(Optional) Opt out of crash reporting

Crewship v0.1 beta enables anonymous crash reporting by default. To opt out before any data leaves your machine:
crewship telemetry off
Full details — what is and isn’t sent, routing to your own Sentry, defense-in-depth scrubbing — in the Telemetry guide.

Create Your First Crew

1

Add a credential

Store your LLM API key in the encrypted vault:
echo "sk-ant-api03-YOUR-KEY" | crewship credential create \
  --name "anthropic-key" \
  --provider ANTHROPIC \
  --type API_KEY \
  --value-stdin
The key is validated against the provider API, then encrypted with AES-256-GCM before storage.
--type API_KEY is for plain provider API keys (sk-ant-api...). If you instead have a Claude Code OAuth token from a flat-rate subscription (sk-ant-oat...), use --type AI_CLI_TOKEN — the runtime injects the two differently, and mixing them up breaks agent authentication.
Use --value-stdin to pipe the key securely. Using --value directly exposes the key in the process list.
2

Create a crew

crewship crew create \
  --name "Dev Team" \
  --slug dev-team \
  --icon code \
  --color blue
3

Create an agent

crewship agent create \
  --crew dev-team \
  --name "Alice" \
  --slug alice \
  --role AGENT \
  --system-prompt "Senior full-stack developer specializing in Go and React."
4

Assign the credential

crewship credential assign anthropic-key alice \
  --env-var-name ANTHROPIC_API_KEY
The --env-var-name flag is required — it specifies the environment variable the agent will use.
5

Run the agent

crewship run alice "Set up a basic Go HTTP server with health check endpoint"
The agent starts in an isolated Docker container, receives your prompt, and streams output in real-time.

Open the Dashboard

Navigate to http://localhost:8080 to see the web UI where you can:
  • View crew boards and agent status
  • Monitor real-time logs from inside each container
  • Manage credentials and skills
  • Create and track missions
  • Chat with agents directly

What’s Next?

Architecture

Learn how the runtime, sidecar proxy, container isolation, and IPC fit together under the hood.

Credentials

Set up credential failover, scoping, and the Keeper guardrail.

Production Checklist

The 6 things a production agent system needs — and how Crewship covers all of them.

Orchestration

Create missions with multi-agent task delegation.