Knowledge Base

What it does

• Ingest PDFs, web pages, YouTube transcripts, plain text, and Google Drive files — all parsed, chunked, and embedded automatically
• Chat with an AI grounded in your own documents — Gemini Flash for speed, Pro for depth, with citations every time
• Semantic search via pgvector — find relevant content even when the keywords don't match
• Deep research mode runs async multi-source workflows that fill gaps in your knowledge and ingest results back into the notebook
• Built-in MCP server exposes your knowledge base to Claude Code and other AI agents — search, chat, and query directly from your tools
• Synthesise AI responses back into the source library, so today's answer becomes tomorrow's context

How it works

1. 1

   Collect

   Drop in PDFs, paste a URL, link a YouTube video, or import from Google Drive. Each source is chunked into 500-token segments and embedded with Gemini.

2. 2

   Chat

   Ask questions in natural language. The assistant runs semantic search, list queries, and (optionally) live web searches, then answers with the sources it actually used.

3. 3

   Research

   Deep research mode identifies the gaps in what you've collected and runs async workflows to fill them — results are stored as new sources you can chat with.

4. 4

   Connect

   The built-in MCP server makes your knowledge base a first-class tool inside Claude Code and other MCP-compatible agents.