Docs · Overview
How ShipFlow works
ShipFlow AI is an AI code-review platform with an autonomous shipping pipeline. It takes a feature request all the way to a reviewed, merged pull request — and keeps a human in the loop at the gate that matters.
The core loop
Every feature request moves through the same, always-visible pipeline. Each stage is durable: if a model call fails, only that step retries — never the whole run.
- 1
Request
Paste a raw request — a customer email, a support ticket, or a one-liner. ShipFlow detects possible duplicates and asks clarifying questions when context is missing. - 2
PRD
A structured product spec is drafted automatically: problem statement, goals, non-goals, user stories, acceptance criteria, edge cases, and success metrics. You can edit everything before approving. - 3
Tasks
The approved PRD is broken into an actionable task board so the work is scoped before any code is written. - 4
Build & AI review
As pull requests open, ShipFlow reviews them with full-repository context — flagging real bugs, security issues, and regressions, then proposing concrete fixes. - 5
Human approval & ship
AI does the review; a human does the deciding. Once approved at the gate, the change is routed to merge and the request is marked shipped.
What makes the review different
Tools like CodeRabbit, Qodo Merge, Greptile, and Korbit comment on a PR after it exists. ShipFlow reads the diff and the surrounding codebase, so reviews understand how a change interacts with the rest of your repo — not just the lines that moved.
Repo-aware by design
Architecture in one minute
The dashboard is a Next.js App Router UI. Actions trigger durable background workflows that fetch from GitHub, call your chosen models, and persist results to Postgres. Because work is event-driven, the UI responds instantly while the heavy lifting happens in the background.
Human-in-the-loop, always