Codeceipt

Ship AI code. Not AI slop. A public, independently-verifiable receipt that proves an AI-authored pull request actually did what it claimed — checked by execution, not by trusting the agent's self-report.

The full brief, pitch, and research live in docs/:

• docs/README.md — product SSOT (positioning, wedge, architecture)
• docs/PITCH.md — Builders Day pitch + booth talking points
• docs/RESEARCH-ANCHORS.md — research grounding

Stack

Next.js 16 · React 19 · TypeScript (strict) · Tailwind 4 · Vercel KV · Zod.

Run it

pnpm install
cp .env.example .env.local   # optional — defaults run with zero config
pnpm dev

Open http://localhost:3000, paste a GitHub PR URL, get a Receipt at /r/<id>.

With no env set: the engine returns a deterministic mock verdict and receipts are stored in-memory. Production wires the real engine (CODECEIPT_ENGINE_MODE=cli) and Vercel KV.

Architecture

PR URL / diff ─▶ /api/scan ─▶ lib/github (fetch diff)
                            ─▶ lib/engine (runEngine: mock | cli)   ◀── swap point
                            ─▶ lib/store  (Vercel KV | in-memory)
                            ─▶ /r/[id]    (public Receipt page)

The web app only ever calls runEngine() — it never knows whether the verdict came from the mock or the real codeceipt-engine CLI. Extracting the engine from cortex-x is a drop-in: implement the diff in → JSON Verdict out contract in lib/engine.ts and set CODECEIPT_ENGINE_MODE=cli.