SHIP runs AI agents that plan, build, review and test issues, then open a pull request. Free credits, packs from $49. See features, pricing and pros and cons.
Researched and fact-checked by Bartu Cavusoglu · Updated October 26, 2026
SHIP is an agentic engineering platform from SHIP AI. It takes an assigned issue through planning, building, review and testing, then delivers a pull request with evidence that the change works. The company describes it as an autonomous software team for engineers, solo founders and engineering teams.
The problem it targets is the gap between how fast AI agents write code and how slowly teams verify it. SHIP adds review and test stages around the code, and records cost for every stage so spending can be tied to delivered work.
You assign an issue by Linear assignment, a Jira ticket, an @SHIP comment on a GitHub issue, or the ship CLI. The Planner reads the ticket, repository, docs and prior pull requests and writes a plan. The Builder codes in a sandbox, runs your formatter, linter, typecheck and tests, and fixes CI failures.
The Reviewer checks the diff against the plan and your project guidelines. Your own pipeline deploys an isolated preview, and the Tester walks the happy and unhappy paths against acceptance criteria. When a gate fails, feedback returns to the Builder. If progress stalls, SHIP stops and asks for you.
The central feature is per-role configuration. Each role can use a different coding agent (Claude Code, Codex, Pi, OpenCode or Kilo Code) and model, set in ship.yml or overridden per mission. Checkpoints let you pause after the Planner, Builder or Reviewer, and a person accepts finished work unless auto-merge is enabled.
Mission Control shows each mission's timeline, turns, duration and cost by stage. Scheduled missions handle dependency upgrades and security patches. A CLI, hosted MCP server and HTTP API let assistants and scripts send work. Agents run in an isolated sandbox where your provider keys are never exposed.
SHIP is an agentic engineering platform that takes an assigned issue through plan, build, review and test using four AI agents. It then opens a pull request with proof it works. It is aimed at engineers, solo founders and engineering teams.
SHIP charges per mission in credits. Packs cost $49 for 10 credits, $199 for 50 and $349 for 100, all one-time. Starter is $170 a month for 50 credits and Scale is $480 a month for 160. Enterprise starts at 12,000 credits a year.
Yes. Connecting your first repository gives 3 free credits, once per GitHub account, with 1 project and 1 agent at once. No credit card is required.
Yes. Agents run on your own model provider account, and your provider bills that usage at your own rates. SHIP's credits pay for its agents and sandboxes.
SHIP works with GitHub, Linear and Jira, and a project can run without a tracker through the CLI. It supports Claude Code, Codex, Pi, OpenCode and Kilo Code, with models from Anthropic, OpenAI, Z.ai and OpenRouter.
Not by default. A person accepts the finished work, and checkpoints can pause the pipeline after the Planner, Builder or Reviewer. Auto-merge can be turned on per project or per mission.
Agents run in an isolated sandbox and do not receive your keys. SHIP says it does not keep a copy of your repository. Session logs, which can include code snippets, are kept for 1 year and deleted sooner on request.
SHIP suits engineering teams and founders with a backlog of well-scoped issues who want review and testing before a person looks at the code. Teams that want to control model choice and cost per stage will find the most value.