BenchLoop Web

BenchLoop is the local AI intelligence network: ask what will work on your hardware, reproduce an exact recipe with the local runner, publish a structured benchmark receipt, and learn from the builders reproducing it.

The product loop is:

Ask → Recipe → Runner → Verified run → Community signal → Better Ask

Surfaces

Directory	Role
site/	Public React app: discovery, profiles, discussions, recipes, runs, and leaderboards
worker/	Cloudflare Worker, D1 migrations, Better Auth, Ask Loop, community APIs, and Runner pairing
api/, ui/	Existing local runner control plane; preserved while the public network is built

The benchmark engine and CLI live in the separate outsourc-e/bench-loop repository.

Public site development

cd site
npm install
cp .env.example .env.local
npm run dev

The site talks to https://api.bench-loop.com by default. Set VITE_API_URL only when developing against a local Worker. Public discovery falls back to the polished launch dataset while a new D1 community has no posts.

OAuth, D1, and Hermes credentials stay in Worker secrets. Never place a secret in a VITE_* variable.

Verification

cd site
npm run check
npm audit --audit-level=moderate

The database foundation lives in worker/migrations/. Apply it with Wrangler, then deploy the Worker and Cloudflare Pages site as described in site/DEPLOY.md.

Product direction

See docs/REVAMP-VISION.md for the product thesis, information architecture, trust model, and milestones.