Crewship
A self-hosted runtime for coding agents: each crew in its own Linux container, wrapped in a control plane with roles and audit
High risk
We rate an entry high when the tool writes to external systems, handles money, production databases or secrets, or runs arbitrary commands. The CLI installs it only with your consent.
Why this level
- Runs coding agents in Linux containers that execute arbitrary code
- Stores and proxies keys and secrets, and manages the host's container runtime
- Open beta: some safety layers are early or off by default
Install
Manual install
brew install crewship-ai/tap/crewshipmacOS and Linux. Then crewship doctor and crewship start, with the UI on http://localhost:8080.
This is third-party code. Review the repository files before installing.
What it does
Crewship runs coding agents on your own hardware, giving each crew an isolated Linux container where the agent can install packages, databases and tools. A crew can be driven by Claude Code, OpenCode or a local model via Ollama, so you are not tied to one provider. Around the runtime sits a team-shaped control plane: missions with a task breakdown, scheduled routines, an issue tracker with triage, role-based access from OWNER to VIEWER, an append-only audit journal and keys in an encrypted vault. A governance layer called Keeper can refuse a secret to an agent and watch behavior after the fact. All code, data, conversations and memory stay with you and pack into encrypted backups. The product is a single Go binary with an embedded Next.js web UI and a full CLI.
Who it is for. For developers, devops engineers and founders who need to run a fleet of agents on their own infrastructure under roles and audit.
Good fit when
- You want to run agents in isolated containers on your own hardware, not someone else's cloud
- You need roles, audit and a key vault around a team of agents
- You want to drive an agent on a local model via Ollama without an external API
Not a fit when
- You need a single agent in an editor without your own infrastructure and containers
- You cannot stand up a host with a container runtime, or you need a stable release rather than an open beta
Example request
Stand up an engineering crew on Claude Code and start a mission to refactor authenticationLimitations
The project is an open beta: some features are marked early or unfinished, and APIs can change between minor versions, so pin a tag for production. It needs a host with a container runtime, runs on a single host, and multi-host scheduling is still planned. The network between crews is not isolated. The tuned engine is Claude Code, which calls the Anthropic API that is not reachable from Russia without a VPN, while fully local work is possible through Ollama. Storage is SQLite, with PostgreSQL planned.
How to disable. Stop the daemon with crewship stop, or docker compose down for the Compose deployment. Remove the binary (brew uninstall crewship for a Homebrew install) and the data directory ~/.crewship/.
Security check
- Runs coding agents in Linux containers that execute arbitrary code
- Stores and proxies keys and secrets, and manages the host's container runtime
- Open beta: some safety layers are early or off by default
README in short
The README presents Crewship as a self-hosted runtime where each crew of agents gets its own Linux container, with a team-grade control plane built around them. The engine is Claude Code, OpenCode or a local model via Ollama. It lists missions, routines, an issue tracker with triage, role-based access, a hash-chained audit journal, a key vault, the Keeper layer and encrypted backups, honestly labeling what is ready, early or in progress. It installs via Homebrew, a signed installer or Docker Compose, starts with crewship start, and has a full CLI with parity to the REST API. A single Go binary with an embedded Next.js UI and a SQLite database. Apache 2.0 license.
FAQ
Do code and data leave my machine?
No, everything stays on your hardware: code, data, conversations, memory and the audit journal, plus encrypted backups. Only the agent itself reaches out, when its engine calls an external model.
Can I avoid a cloud model?
Yes, a crew can run on a local model via Ollama under the OpenCode adapter, with no external API. Claude Code, by contrast, always calls the Anthropic API.
Related
Open-source personal AI assistant on your own machine: answers in Telegram, Slack, Discord and WhatsApp, extended with skills and plugins
A self-improving agent from Nous Research with a TUI, messaging gateway, cron jobs and skills it writes itself
An open source coding agent for the terminal and desktop with build and plan modes
Open prompt library with a Claude Code plugin, MCP server and CLI: search, fetch and improve prompts and skills from an agent