Shipfox: Your AI software factory

Shipfox is your AI software factory. Build workflows that run shell and AI agent steps, start them from the tools you already use, and run them on compute you control.

Your AI software factory.

Shipfox turns your engineering work into workflows that AI agents can run. You write each workflow as YAML in your repository. A matching event can start a run, such as a code push or an alert. An agent can read context, use selected tools, and act within the workflow's permissions. Every step streams to the dashboard as it runs.

See how it works

The run model in one page: triggers, jobs, steps, agents, and runners.

Get started

Connect a repository and run your first workflow in a few minutes.

Anatomy of a workflow

A workflow starts with a trigger, then runs one or more jobs. Each job is an ordered list of steps, and a step is a shell command or an AI agent. Larger workflows chain many jobs and run them in parallel.

This fragment adds a Sentry trigger and a job below the top-level name and runner fields of an existing workflow:

triggers:
  on_alert:                        # Define any trigger from your integrations
    source: sentry_prod              # Replace with your Sentry integration connection slug.
    event: issue.created
jobs:
  fix:
    steps:
      - run: npm install           # a shell step
      - prompt: |                  # an agent step uses workspace defaults
          Investigate the alert in the repository and report the likely cause.
          Do not change files.

Agents automate your process. Any step can be an AI agent that reads your repository, runs tools, and acts on the result.
Watch every run. Runs, jobs, and steps stream live to the dashboard. Agent steps show their messages, thinking, tool calls, tokens, and cost. See Workflows and runs.
Works with your daily tools. Start runs from GitHub, Sentry, or any custom webhook. See Integrations.
Your harness, your models. Pick your agent harness and any of the supported model providers.
Run on your compute. Jobs run on isolated runners you control. Use long-lived capacity or provision a runner for each job.
Open source, MIT. Self-host the whole platform. Shipfox on GitHub.