Book a demo
SHIP in your organization
Schedule a conversation to see how SHIP can help you reduce your AI costs by enabling your teams to turn tickets into reviewed, verified work with agent fleets that operate inside your existing delivery process.
- Autonomous deliveryOperate AI agents that review and test as fast as they build, so work ships while your engineers stay in control.
- Cost optimizationMonitor every step, compare performance, and optimize AI costs confidently.
- AI sovereigntyRun missions on the models, tools, and AI contracts your organization already uses.
Pi
How SHIP fits into your organization
Read what SHIP runs today, what a cycle costs, and what stays under your control.
How is this different from a coding assistant?
An assistant helps an engineer write code. SHIP takes an assigned issue through plan, build, review, deploy, and test, then reports what it delivered and what it cost.
How is this different from vibe coding?
Vibe coding leans on casual prompts and skips the stages that catch mistakes, which works until the change has to survive a review. SHIP runs the same work through a plan, your CI, a reviewer and a test stage, and loops back with the feedback when a gate fails. Because the model is set per role, you can put a cheap one on the first pass and a stronger one on the review that judges it.
Which coding agents and models can SHIP run?
SHIP supports Claude Code, Codex, Pi, OpenCode, and Kilo Code as agent harnesses. For models, it supports Anthropic, OpenAI and Z.ai directly, plus every model available through OpenRouter. You can assign a different harness and model to each role, from budget to frontier, without changing code.
Do I need an issue tracker?
No. Linear and GitHub Issues both work if your team already plans there, and a project can run with no tracker at all, in which case you hand work over from your terminal with the CLI and the missions live in the console. Either way the work lands as a pull request in your repository.
Can I use my active subscription keys?
Yes, as long as your provider allows programmatic access with your subscription keys. SHIP orchestrates agents just like you do in your local environment. If you're able to use your keys for non-interactive sessions on your CLI, you're good to go. Subscription OAuth tokens typically have 1 year of lifetime.
What does a cycle cost?
It depends on the work and the models behind it. Every cycle reports its own cost, attributed by stage, priced on the model that actually served each request. Our SWE-in-a-team benchmark put 260 runs through the full loop on a private fixture: a clean run came in at a median of $3.39, and the same twenty tickets cost $67.11 on one builder against $122.42 on another.
Can I change a plan before the build starts?
Yes. Put a checkpoint on the planner and the pipeline holds as soon as the plan is ready, before any code is written, so you can read it and send follow-up instructions. Checkpoints, the harness and model per role, and the rest of a project's defaults all live in one ship.yml at your repository root.
Do I have to approve every change?
By default a human accepts the work at the end. Put a checkpoint at any stage to hold the pipeline there and read what it produced, or turn auto-merge on per project or per mission for work where you have decided you do not need that gate.
What happens when an agent gets it wrong?
Review and test stages block the work, and the loop retries with that feedback. When it stops making progress, it stops and asks for you.
Is SHIP cross-ticket aware?
No. Each mission plans and builds on its own branch, and nothing coordinates two missions that touch the same files, so keep concurrent issues small and separable or have one agent sequence them. If two branches do collide, SHIP spots the conflict before it waits on CI and puts the builder back on the branch to resolve it, under the same retry budget as any other failure.
Do tests run against production data or a preview environment?
Neither is a copy of production. By default tests run against a preview environment deployed for that pull request by your own pipeline, with any database it needs created from your migrations and seeded from committed fixtures. Seeding from a live environment is possible but opt-in, and never from production. A project with nothing to preview, a CLI for instance, is built and exercised locally instead.
Does any agent see my secrets?
No. Agents work in an isolated sandbox with scoped, per-task access. The platform holds your credentials and calls providers on the sandbox's behalf.
Does SHIP keep a copy of my code?
No. Your repository is cloned into an isolated, ephemeral sandbox that is destroyed when the run ends. SHIP keeps session logs of what each run did for observability, but not a copy of your repository.
Can it run on my own cloud and CI?
Yes. Deploy and test run through your own SDLC, including the CI and deployment pipelines you already have. Inference runs on your own provider contracts. Agent execution runs in a secure sandbox, and the session data that powers your logs, metrics, and audit trail is held under SHIP's own security controls.