Chat with your PDFs.
Answers you can cite.
Quill is a retrieval-augmented chatbot template. Upload a document, ask grounded questions, and get answers backed by exact source citations — with an explicit fallback when it doesn't know.
Everything a docs chatbot should have
Built for the "AI chatbot over my docs" job — and structured so you can ship it to a client with confidence.
Cited answers
Every claim is grounded in numbered source blocks pulled from your PDFs — no hallucinated references.
Honest "I don't know"
When the retrieved context doesn't support an answer, Quill says so instead of making things up.
Upload & chunk
Drop in a PDF; text is extracted, split into overlapping page-aware chunks, and embedded automatically.
Streaming responses
Answers stream token-by-token over Server-Sent Events for a snappy, live feel.
Separable modules
LLM, embeddings, PDF, RAG, and DB each live in their own folder under lib/. Swap any layer.
One-click summaries
Generate a structured overview of any document with key points and notable details.
How it works
Three steps from a raw PDF to a cited, streaming answer.
Upload your PDF
Pick a policy manual, contract, FAQ, or any document set. Files are stored privately per session.
We index the content
Text is extracted, chunked, and embedded with OpenAI's text-embedding-3-small into pgvector.
Ask grounded questions
Your question is embedded, matched against the index, and the model answers with citations.
Drop in your own domain
Quill is domain-agnostic. Clients we've shipped this to use it for legal, real estate, healthcare, and SaaS docs — the pipeline is identical, only the content changes.
Open the demoLaw firm FAQ
Ground answers in case archives, retainers, and firm policies.
Real estate docs
Query leases, contracts, and listings without reading every page.
Clinic policy manual
Let staff ask procedural questions and get cited policy answers.
SaaS docs
Turn your product documentation into a conversational assistant.
Ship a RAG app this week
A complete, separable template. Configure your keys, deploy to Vercel, and hand it to a client.