| Requirement | When it is needed |
|---|---|
| Node.js 20 or later, with npm | Installing, building, and running. The minimum is in package.json; installation examples use Node 24. |
Git on PATH | Git projects, clones, diffs, and worktrees. |
| Bash and a POSIX environment | Default shell runner (/bin/bash) and full terminal support. Use WSL2 on Windows. |
Python 3 as python3, with POSIX modules | Real terminals without installing node-pty; recommended on Linux, macOS, and WSL. |
| Browser with JavaScript and WebSocket | The web interface. |
| Folder writable by your user | Code, .cockpit/ state, logs, and worktrees. |
| Internet | Dependencies, voice models, plugins, and remote providers. |
| Installed, authenticated AI CLI | AI tasks; a local shell works without a provider. |
Cockpit tries terminal backends in this order: node-pty, the Python POSIX bridge, then pipes. node-pty is not a declared dependency and is unnecessary when the Python bridge works. Pipes provide neither a real TTY nor effective resizing, which may affect interactive CLIs.
#What you do not need
- Docker, PostgreSQL, Redis, or a separate SQLite server.
- A separate Cockpit installation, DeepSeek Harness, Ollama, or Obsidian.
- Every provider, skill, or media tool.
- Global TypeScript, Vite, or React installations: npm installs project tools.
- A GPU for default CPU-based transcription.
No universal API key is required to start the dashboard. Each provider has its own authentication, quotas, and billing. Installing Cockpit does not grant model access.
#What to avoid
- Running the server as root/administrator to work around permissions.
- Mixing Windows Node/npm/CLIs with those installed inside WSL.
- Copying
node_modulesacross systems or architectures; reinstall withnpm ci. - Committing tokens, credentials,
.env, session state, or vault files. - Deleting
.cockpit/or worktrees without a backup to fix an error. - Treating worktrees as security sandboxes: processes retain your user's permissions.
The repository does not specify a validated minimum RAM or disk capacity. Usage depends on your team, projects, plugins, and local models. Start with a few agents and monitor your machine.