Docs · Getting started · Concepts · Slack community · Contributing
Stripe, Spotify, and Uber each built a platform to run their coding agents. Shipfox gives your team the same platform, without building it yourself.
- An event starts a workflow. A ticket, a new pull request, an alert, a failed check, or a schedule.
- The workflow does the work. It mixes AI agents with the exact commands you choose, in an isolated environment with a full log and the cost of every run visible.
- The result reaches your team. A pull request, a diagnosis, or a suggested improvement, ready to approve. Leave a comment and the workflow continues.
Workflows live as YAML in your repo, and you review them like any other code. Run on Shipfox Cloud or on your own infrastructure. Any model per step, any stack. Open source.
A big part of your team's week goes to recurring work: small tickets, failing checks, alerts, dependency updates, release chores. Shipfox finds what to automate, ships it as pull requests you review, and improves it every week.
- Find. Connect your coding agent to Shipfox, and it suggests the workflows that fit your repository and tools.
- Ship. Each workflow arrives as a pull request, tested before you commit. You set which model runs each step, what each agent can touch, and where a person signs off.
- Improve. Workflows watch your other runs, explain each failure, and suggest a fix as a pull request.
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.