Shipfox

Your software factory, smarter every week

Shipfox finds what to automate, ships it as pull requests you review, and improves it every week.

Stripe, Spotify, and Uber each built a platform to run their coding agents. Shipfox gives your team the same platform, without building it yourself.

Run coding agents like CI pipelines

How a Shipfox workflow runs

1. Triggers

Linear, GitHub, Slack, Jira, Sentry, PostHog, Notion, ClickUp, a schedule, or a webhook start the workflow, from a ticket, a pull request, an alert, a check, or a schedule.

2. Agent workflow · isolated, ephemeral runner

Defined in .shipfox/workflows/implement-issue.yml
  1. Triage: agent step on GLM 5.3, reads Linear, costs about $0.01.
  2. Mark in progress: tool step, writes Linear.
  3. Fix issues: agent step on GPT-6 Sol, edits code, costs about $0.12.
  4. npm test: run step. If it fails, the workflow goes back to Fix issues.
  5. Open PR: agent step on Claude Sonnet, writes GitHub, costs about $0.03.

Example run #128 succeeded in 4m 12s, cost about $0.16, and recorded logs, the agent session, and metrics.

The Open PR step sends the result to your team. A comment on it, even days later, resumes the Fix issues step with full context.

3. Your team

Your team gets a pull request marked ready, with tests passing, a summary of changes, and a small, focused change ready for review. Other workflows return a diagnosis or a suggested improvement instead.

  1. An event starts a workflow. A ticket, a new pull request, an alert, a failed check, or a schedule.
  2. 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.
  3. 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 software factory that improves itself

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.

How Shippy AI improves your workflows

Shippy AI watches your workflows, the 9 tools they connect to, and every run. Each change it makes reaches your team as a pull request, and the platform runs the new version once you approve it.

  1. It found 30 Sentry alerts a week triaged by hand and opened PR #41 to add an alert-triage workflow.
  2. After release-notes failed 3 times at the changelog step, it fixed the step in PR #52.
  3. It moved the diagnosis step of alert-triage to a smaller model that passes the same checks, in PR #57, which cut the cost of a run from about $0.42 to $0.12.
  4. It found 12 code questions a week in Slack and opened PR #63 to add an ask-codebase workflow.
  5. It saw pr-checks spend 3 of its 5 minutes installing dependencies and cached them in PR #68, which brought a run down to about 2 minutes.
  • 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.
Was this page helpful?
Edit this page on GitHub

On this page