doc-chat

Chat with your own documents. Upload a PDF, paste text, or point it at a URL, and ask questions. Every answer cites the chunks it came from, and the model is told to say so rather than invent when the sources do not cover the question.

RAG over pgvector, end to end, in about 400 lines.

What it does

1. Ingest. Extract text from a PDF, a pasted block, or a scraped web page.
2. Chunk. Split on paragraph boundaries with overlap, dropping fragments.
3. Embed. Voyage voyage-3, batched, with the document input type.
4. Store. Postgres with pgvector, through Supabase.
5. Ask. Embed the question with the query input type, pull the nearest chunks, and stream an answer grounded in them.

Both the ingest and the chat endpoints stream NDJSON, so the UI shows progress during a long embed and renders citations before the first token of the answer arrives.

Stack

Piece	Choice
Framework	Next.js 16, App Router, node runtime
Embeddings	Voyage voyage-3
Vector store	Supabase Postgres with pgvector
Answering	Anthropic claude-haiku-4-5
PDF extraction	unpdf
Styling	Tailwind 4

Setup

npm install
npm run dev

You need four environment variables:

Variable	For
VOYAGE_API_KEY	Embeddings
ANTHROPIC_API_KEY	Answering
SUPABASE_URL	Vector store
SUPABASE_ANON_KEY	Vector store

The database needs two tables and one RPC before anything works. The schema is in TECHNICAL.md, including the match_chunks function the search path calls.

Layout

app/api/ingest/route.ts   extract, chunk, embed, store; streams progress
app/api/chat/route.ts     retrieve, then stream a grounded answer
app/page.tsx              the whole UI
lib/chunker.ts            paragraph-first splitting with overlap
lib/embed.ts              Voyage client, document and query modes
lib/store.ts              Supabase reads and writes
lib/types.ts              shared shapes, including the stream event union

How it works

TECHNICAL.md covers the chunking strategy and why it cascades through three boundary types, why the document and query embedding calls are separate functions, the database schema and the match_chunks RPC, the NDJSON streaming protocol, and the known limits of the URL scraper.