Fanout

Fanout turns one company website into a managed AI-visibility delivery: buyer-search research, a bounded site audit, retrieval analysis, bottom-of-funnel page drafts, homepage improvements, technical search files, and a reviewable delivery pack.

What this stage includes

• concurrent OpenAI, Anthropic, and Gemini query-research adapters;
• a zero-key deterministic demo replay;
• public-page retrieval, chunking, scoring, and rewrite rescoring;
• bounded same-origin site auditing;
• source-aware growth-pack generation;
• review staging and verification routes;
• the complete landing workflow and guided demo chapters;
• an internal Agency OS command center for engagement queues, work orders, approvals, artifacts, and delivery;
• a Live Build Studio that shows agents rewriting the homepage, preparing technical fixes, and completing a ten-page programmatic SEO queue.

Open /agency to operate and review saved service runs. Switch to Live build studio to see the completed Webform delivery sequence.

Hackathon pitch

The final judge deck is available as PDF and PowerPoint.

Run locally

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

Open http://localhost:3000. The guided demo works without provider keys. Live provider runs use server-only environment variables from .env.local.

Provider configuration

OPENAI_API_KEY=
ANTHROPIC_API_KEY=
GEMINI_API_KEY=

OPENAI_MODEL=gpt-5.5
ANTHROPIC_MODEL=claude-sonnet-5
GEMINI_MODEL=gemini-3.5-flash

The three provider adapters use native HTTP APIs, so no provider SDKs are required. The registered model search tool currently returns a small deterministic result corpus; it captures the models' literal query choices but is not a live search-engine index.

Checks

npm run typecheck
npm run lint
npm run build
npm test

Provider keys and local environment files are ignored and must never be committed.