Docs · Getting started · Concepts · Slack community · Contributing
The agent platform for engineering teams.
Shipfox turns engineering work into automated workflows built from AI agents and shell commands. Workflows live as YAML in your repository and react to events across your stack. Shipfox handles orchestration, secure tool access, isolated execution, and monitoring.
Get your first workflow running now.
- Agents that stay under control. Agent steps, where a model decides what to do, sit next to shell steps that run your exact commands. You set the structure. The model works inside it.
- Loops that run until the result is correct. A gate is a pass/fail check on a step. When it fails, the workflow loops back to an earlier step and tries again, up to a safe limit. That is how an agent keeps going until the tests pass, with no scripting.
- Long-running, event-driven agents. A listening job stays alive across a run and runs an agent on each new batch of events (PR review comments, new issues) until a resolution condition is met. Asynchronous agent loops, not one-shot runs.
- Triggers from your whole stack. Start runs from GitHub, Sentry, Slack, Linear, and more through integrations. Missing one? Point it at the generic webhook and trigger on its events too. Connect several of the same provider and target each independently.
- Secure by design. Each job runs isolated in a runner next to your code that polls outbound for work, so nothing connects in. No data stays between two runs, and each agent reaches only the tools you allow.
- One place to control everything. Every run streams its jobs, steps, agent messages, thinking, tool calls, tokens, and cost while it happens.
- Your harness, your keys. Run an agent step on the
piharness (any of 30+ model providers) or theclaudeharness (the Claude Agent SDK on your Anthropic key), chosen per step. - Open source and self-hostable. The whole platform is MIT licensed. Run it on your own infrastructure so that your code and your credentials never leave it.
- Triage monitoring errors. A new error starts an agent that produces a fix and opens a pull request.
- Turn tickets into code. An assigned ticket starts an agent that opens a pull request and responds to review comments.
- Fix failing CI. A failing check starts an agent that produces a fix and makes the check pass again.
- Review pull requests. Each new pull request gets an agent review based on the team's guidelines, with follow-up on later changes.
Start with the Getting Started guide. Self-hosting Shipfox? See the installation docs.
Contributing to Shipfox? Read CONTRIBUTING.md.
Full documentation is published at shipfox.io/docs.
- Slack: join the Shipfox community Slack for help, ideas, and release news.
- Issues: report bugs and request features on GitHub Issues.
Report vulnerabilities privately by emailing security@shipfox.io rather than opening a public issue. See SECURITY.md for details. Shipfox's token and trust model is documented in the auth security model.
Read CONTRIBUTING.md before you open a pull request. Its task map links to the engineering guide for each change.