A production-ready MVP for document chat with Next.js 14, Supabase pgvector, and the raw OpenAI SDK. Upload PDFs, index them, and chat with streaming answers that include source citations.

• PDF upload -> chunk -> embed -> store in Supabase pgvector
• URL import/crawling with automatic content extraction
• Streaming chat responses with markdown rendering and source citations
• Feedback-driven retrieval — thumbs up/down votes adjust chunk quality scores to improve future results
• Admin monitoring dashboard with feedback rate, positive rate, knowledge gaps, and chunk usage stats
• Clickable source badges (links for crawled URLs, expandable content for uploads)
• Clean minimal light-mode UI using Tailwind + shadcn-style components

• Next.js 14 App Router + TypeScript
• Tailwind CSS + shadcn/ui-style components
• Supabase Postgres + pgvector
• OpenAI SDK (no LangChain)

npm install

Create a Supabase project and run the SQL migrations in order:

# Run each migration file in the Supabase SQL Editor:
supabase/migrations/0001_init.sql        # Base tables, pgvector, match_chunks
supabase/migrations/0002_add_url_source.sql   # URL source tracking on documents
supabase/migrations/0003_match_chunks_url.sql # source_url in match_chunks results
supabase/migrations/0004_feedback_quality.sql # quality_score, feedback-driven retrieval, chunk usage stats

Create a .env.local file:

OPENAI_API_KEY=sk-...
SUPABASE_URL=https://YOUR_PROJECT.supabase.co
SUPABASE_SERVICE_ROLE_KEY=YOUR_SERVICE_ROLE_KEY

Notes:

• The app still runs without env vars. Upload + chat will return helpful warnings.
• Use the Supabase service role key for server-side writes.

npm run dev

Visit:

• http://localhost:3000 - main app
• http://localhost:3000/admin - admin dashboard

• POST /api/upload - Upload PDF and index chunks
• POST /api/crawl - Import and index content from a URL
• GET /api/documents - List uploaded documents
• POST /api/chat - Streaming chat with markdown and citations
• POST /api/feedback - Thumbs up/down (also updates chunk quality scores)
• GET /api/admin/stats - Dashboard stats, knowledge gaps, chunk usage
• GET /api/admin/queries - Recent queries

1. Push the repo to GitHub.
2. Create a new Vercel project and import the repo.
3. Add the environment variables in Vercel project settings.
4. Deploy.

If you want to self-host, ensure the same env vars are available at runtime.

• app/ - Next.js App Router pages and API routes
• components/ - UI components and client UI
• lib/ - OpenAI, Supabase, chunking, embeddings helpers
• supabase/migrations/ - SQL migrations for pgvector tables and RPCs

MIT