🧠 notebooklm-agent

Zero-token AI agent powered by Google NotebookLM.

No API keys. No external LLMs. NotebookLM's built-in Gemini does all the reasoning.

Features

• Zero external tokens — NotebookLM's Gemini is free
• Brain-first architecture — every notebook auto-bootstrapped before use
• Atomic research pipeline — brain_research() does start→poll→import→wait
• Persistent memory — notebooks grow smarter over time
• Multi-gateway — Telegram, Discord, CLI, WebSocket
• Custom ReAct loop — no smolagents dependency, direct chat.ask()
• Async-native — no sync/async bridge hacks

Install

pip install notebooklm-agent

Quick Start

# Login to Google (opens browser for OAuth cookies)
nlm-agent login

# Chat with a new notebook (auto-bootstrapped + auto-researched)
nlm-agent run "Explain quantum computing"

# Start Telegram bot
nlm-agent serve --gateway telegram

# Research a topic
nlm-agent research "fusion energy 2025"

# Generate podcast from research
nlm-agent podcast

Architecture

User → Gateway (Telegram/CLI/Web)
         ↓
      Agent Core (ReAct loop)
         ↓
      Brain (NotebookLM chat.ask)
         ↓
      notebooklm-py (API client)

Every notebook gets bootstrapped with agent instructions before first use. Every question goes through Gemini via chat.ask() — no local model needed.

License

Apache-2.0