Skip to main content
Shipfox is a continuous shipping platform where any step in a workflow can be an AI agent — one that reads your repository, runs tools, and acts, with its reasoning streamed live — right beside the shell commands, triggers, and runners you’d expect. Workflows are YAML in your repo, and agent steps share the same lifecycle as shell steps: they stream output, report status, and can be gated and retried until the result is correct. This documentation covers everything from connecting your first project to authoring agentic pipelines that fix code and verify themselves.

Get Started

Connect your first project and run a workflow in minutes.

​
What is Shipfox?

Workflows live in .shipfox/workflows/ as YAML files, versioned and reviewed alongside the rest of your code. A trigger — such as a GitHub push, a Sentry alert, or an on-demand call — starts a run. Each run is composed of jobs, which are logical groups of steps. A step is either a run command (plain shell) or an agent step that pairs a model with a prompt. AI agents are first-class citizens in Shipfox, not an afterthought bolted on later — any step in any job can invoke an AI model from 30+ supported providers.

​
Why Shipfox is different

Three things set Shipfox apart, and they compound:
  • Agents are first-class steps. Any step can be an AI agent — pick a model, write a prompt, choose a thinking depth. No plugin, no action, no glue.
  • Agents iterate until the work is correct. Pair an agent with a gate: the agent makes a change, a run step verifies it, and restart_from loops back to the agent until the check passes — bounded automatically.
  • You see the agent think. Agent steps stream structured events — messages, thinking, tool calls, token usage, and cost — not just raw text.
jobs:
  fix:
    steps:
      - key: edit
        model: claude-opus-4-8
        prompt: Fix the failing unit tests in this repository.
      - run: npm test
        gate:
          success_if: exit_code == 0
          on_failure:
            restart_from: edit   # the agent tries again until tests pass

​
Key features

YAML Workflows in Your Repo

Workflows live under .shipfox/workflows/, versioned and reviewed like any code.

AI Agent Steps

Any step can run an AI agent: pick a model, write a prompt, and Shipfox does the rest.

Gates & Retry Loops

Gate a step on its exit code and restart_from an earlier step — bounded loops, no scripting.

Live & Agent Logs

Tail every step live; agent steps show messages, thinking, and tool calls as structured events.

Your Own Runners

Jobs execute on runners you register and control — no vendor lock-in on compute.

Autoscaling Provisioners

A provisioner starts ephemeral, single-job runners on demand and lets them exit when idle.

Smart Triggers

GitHub pushes, Sentry issues, webhooks from any system, or on-demand fire — with more integrations on the way.

30+ Model Providers

Anthropic, OpenAI, DeepSeek, xAI, Google, Mistral, and more.

​
Quickstart

1

Connect your project

In your workspace, create a project and connect a Git repository. Shipfox will watch the repo for the triggers you configure in your workflow files.
2

Add a workflow file

Create .shipfox/workflows/hello.yml in your repository with the following minimal workflow:
name: Hello world
runner: ubuntu-latest
triggers:
  on_demand:
    source: manual
    event: fire
jobs:
  build:
    steps:
      - run: echo "hello world"
3

Register a runner

Generate a registration token in the Shipfox dashboard, then start a runner on your own infrastructure using the matching label. The runner will connect and begin polling for jobs.
4

Fire your first run

Trigger the workflow by clicking Run next to it in the dashboard, or push a commit to your connected repository. Watch the run execute live in the Shipfox UI.
These docs are AI-ingestible — every page is available as Markdown, so your coding agent can author workflows for you.