Crew44

What does this agent do, and when should you use it?

Crew44 is a local-first desktop orchestrator that turns existing coding-agent CLIs into a coordinated crew. An Electron + React 19 renderer talks over WebSocket JSON-RPC to a Go daemon bound to 127.0.0.1; the daemon discovers installed runtimes (Claude Code, Codex, Cursor Agent, Gemini CLI, OpenCode, Qwen Code and more), routes turns to them as subprocesses, and persists everything as plain files under ~/.crew44/. You define Agents — a persona bound to one runtime and model with an instruction and attached skills — and skills are file-based SKILL.md bundles injected into the session across providers. Handover markers pass a turn to a teammate with a one-line brief, optional Worktrees run a task on its own crew/… git branch, and an optional Goal locks a criteria checklist that an independent verifier in a fresh session must confirm. The default crew ships with a Partner, an Engineer, a Product Lead and a Designer. A companion Expo app pairs over an end-to-end encrypted Noise tunnel via a self-hostable relay.

On launch the Go daemon scans disk for installed coding-agent CLIs and exposes runtime discovery, agent/skill/chat state, and an event stream over WebSocket JSON-RPC, authenticated with a per-launch bearer token (only /health is open). In the Crew view you bind each Agent to a runtime, model, instruction and skills folder; a Skill is a SKILL.md plus assets stored in one shared directory that works across every provider. Starting a chat spawns the chosen CLI, streams turn events into an append-only events.jsonl, and runs one in-flight response at a time. Agents emit Handover markers to pass the baton with a short brief. Toggling Worktree gives the chat an isolated git checkout on a crew/… branch without touching your working tree; setting a Goal makes the crew scope a locked criteria checklist and requires a fresh-session verifier to confirm each item before sign-off. The renderer covers Crew, Tasks, New Task, Auto suggestions and Onboarding; daemon config comes from HOST/PORT, AUTH_TOKEN and CREW44_STATE_DIR.

1. Developers who already run Claude Code and Codex and want planning, coding and review done by different models
2. Teams that need project memory and shared SKILL.md workflows so conventions persist across tasks
3. Anyone who wants an agent to complete and verify a task on an isolated branch without disturbing the main working tree
4. Quality-sensitive work where an independent verifier must confirm a locked checklist before you accept the result
5. Mobile users who want to read or nudge a running local crew from their phone

How do you install or deploy this agent?

Signed desktop builds are offered free with no account:

macOS (Apple Silicon)  .dmg
Windows (x64)          .exe installer
Linux (ARM64)          .AppImage / .deb

Downloads are at crew44.io/download. To build from source you need macOS, Windows or Linux, Node 20+, Go 1.26+, and at least one coding-agent CLI installed:

npm install
npm run dev

That builds the Go daemon, starts Vite, launches Electron and connects the renderer once the daemon reports healthy.

How do you use this agent?

Configure the daemon with environment variables:

HOST=127.0.0.1 PORT=8080 AUTH_TOKEN=<token> CREW44_STATE_DIR=~/.crew44 npm run dev

With auth enabled, /rpc requires WebSocket subprotocols:

new WebSocket("ws://127.0.0.1:8080/rpc", [
  "crew44.rpc.v1",
  `crew44.bearer.${token}`,
])

In the app, confirm the detected runtimes on the Crew page, bind models and skills folders to the Partner, Engineer and other agents, then create a task from Tasks with optional Worktree or Goal settings. Other scripts:

npm run build    # renderer build + local app
npm run web:dev  # daemon + bare Vite dev server
npm run test     # Go, renderer and mobile tests
npm run clean    # remove local build artifacts

What are this agent's strengths and limitations?

How does this agent compare with similar options?

Key facts side by side with the most closely related agents.

How does FollowAgents rate this agent?

FAQ