Skip to main content

RAG feature / NotebookLM clone within BoltAI

Feature RequestsHigh Priority

Hey Daniel,

I was thinking about what we should tackle next, and I had a pretty cool idea. You know that Google NotebookLM app, where you can basically upload your PDFs and "talk" to them? We should seriously consider add this as part of BoltAI.

Why this is necessary:

  • No More Guesswork: The AI stops "making things up" (hallucinating) because it's tied directly to the user's own truth.

  • The Ultimate Personal Assistant: BoltAI becomes a personal expert, a research assistant, and a project manager all in one. It turns a "cool" AI tool into an "can't-live-without-it" part of their workflow.

What do you think?


Cheers

Max

Posted by unmatchedintel ·
Status: In Progress6 comments

Log in to comment and vote

Comments6

  • Daniel Nguyen

    Team•

    Oct 9

    Pinned

    Hi all, an update on this one.

    BoltAI 2 can already work from your own files. In a project, open Project Settings… → Linked Sources and link a folder. From then on, any chat in that project can search the folder and open the files it needs to answer, so you don’t attach them one by one.

    I chose this over a classic RAG setup with a vector index for now. It’s the same approach Claude Code and Codex use: search first, then read the real document. Answers come from the full text and always from the current version of the file, with no indexing to wait for or keep up to date. How it works, and the trade-offs, are here: https://help.boltai.com/en/articles/6099742-projects-organize-chats-and-work-with-your-own-files

    Today it reads text, Markdown and code files. PDF, Word, Excel and PowerPoint files in linked folders are next. Until then, PDFs work as project Attachments.

    @Olivier Sarrouy , @Mandy Steinhardt : your notes folders should work right now. Could you link one and tell me how it goes? I’d most like to hear about questions where it didn’t find the right file. Mandy, it should also use fewer tokens than reading whole files through MCP, since it only sends what it reads.

    @ikum : sorry for the slow update.

    • Mandy Steinhardt

      •

      Oct 24

      This is working well! I think better than the MCP setup I had for my files.

  • Mandy Steinhardt

    •

    Aug 30

    Hrm, I use the Bolt AI for research and synthesis and so far I have dealt with this by using my MCP to point to folders that contain the info I want evaluated. However I wonder if a specific RAG setup would enable you to accomplish this with lower input token levels? Right now that is my main problem, trying to reduce the token/api model spend on synthesis, or getting it tuned in properly.

  • TomHsiung

    •

    Mar 29

    Hello, guys

    This is a great idea and I use NotebookLM a lot. I’m a healthcare provider and NotebookLM helps me read reference via the deep dive audio feature.

  • ikum

    •

    Mar 12

    @Daniel Nguyen - any update on this feature/suggestion?

  • Olivier Sarrouy

    •

    Nov 30, 2025

    Really love the idea and would love pushing forward.

    I’m an academic / researcher and this kind of feature would real make a difference. This could literally replace reference / pdf manager, NotebookLM and many more in my workflow.

    I know that I also have tons of notes / markdown files where I randomly push ideas, references and all and it always ends up as a mess to organize and browse. If it could automatically sync with a note folder and something like to integrate it and been able to query it, that would be a second brain and couldn’t live without anymore.

    Cheers.

    Olivier