RAG-powered research, grounded in your sources

Chat with your documents.
Every answer cited.

ChaiBookLM is an AI research assistant that turns PDFs, websites, and videos into a notebook you can actually talk to — with traceable, source-backed answers.

Notebook · “RAG Primer”
How does retrieval-augmented generation work?

RAG combines retrieval with generation. Documents are chunked and embedded into a vector store, then the question is embedded and the nearest chunks are retrieved by similarity before the model answers. 1

[1] Introduction to RAG

Built for trustworthy research

Production-grade architecture with the features that matter for grounded, citation-first answers.

Isolated notebooks

Each notebook is its own knowledge base. Conversations never leak across your projects.

Every answer cited

Responses point back to the exact page, section, or timestamp they came from.

Streaming answers

Watch grounded responses generate in real time with markdown and code rendering.

Query rewriting

A lightweight model rewrites vague questions into retrieval-friendly queries first.

Semantic search

Dense embeddings in Qdrant surface the most relevant chunks for your question.

Grounded, never guessing

If the context doesn't contain an answer, the assistant says so — no fabrication.

From upload to insight

A clean, modular RAG pipeline you can follow end to end.

01

Upload your sources

Drop in PDFs, paste a website, or link a YouTube video. Each source is parsed and chunked.

02

We index everything

Chunks are embedded and stored in a vector database, ready for semantic retrieval.

03

Ask, get grounded answers

Chat with your notebook and receive streaming answers with clickable citations.

Bring any source

Five source types, one retrieval pipeline.

PDFPapers, reports, books
Plain TextNotes and drafts
WebsiteArticles via Jina Reader
YouTubeVideo transcripts
VTTCaptions & transcripts

Start your first notebook

Create a notebook, add a source, and ask your first grounded question in under a minute.

Get started