Documentation
How the stack is put together, how to call the API, and how to run it yourself. The source of truth is the repo — these pages are summaries with runnable examples.
What's running behind this site
Server components where possible; client components only for the search bar, results, and filter sheet.
Express handler bootstraps Nest once per cold start and reuses the app across warm invocations. Swagger at /api/docs.
Pooled connection (pgbouncer, port 6543) for runtime queries; direct (5432) for Prisma migrations. HNSW-ready vector column.
768 dimensions via outputDimensionality. RETRIEVAL_DOCUMENT at ingest, RETRIEVAL_QUERY at search.
Workers skipped in serverless mode; ingestion runs from a laptop or a Vercel cron POST against a trigger endpoint.
One JSON line per request. Health endpoint reports DB, Redis, and Gemini reachability separately.
Guides
API Reference
RESTREST endpoints: /search/hybrid, /search/semantic, /search/text, /search/suggestions, /search/stats. Each with request/response shape and a curl example.
Integration Guide
How-toCalling the deployed API from a client app, wiring the JS SDK shim, running the full stack locally with Docker Compose.
Run it locally
git clone https://github.com/Shailesh93602/CodeSenseiSearch.git
cd CodeSenseiSearch
docker compose up -d # Postgres (pgvector) + Redis
pnpm install
pnpm --filter api db:migrate
pnpm dev # web on :3000, api on :3001Full setup including the Gemini API key and other env vars is in the repo's README.