Notebook Llama: The Open-Source Notebook LM Alternative for AI-Powered Podcasts

What is Notebook Llama?

Notebook Llama is a fully open-source document-to-audio tool developed by the run-llama community, designed as an alternative to Google’s Notebook LM. It enables users to locally deploy an AI system that transforms documents like PDFs into podcast-style spoken content. By leveraging LLaMA models and open-source TTS engines, it creates natural, multi-speaker dialogue from written material—ideal for document exploration, podcast creation, and privacy-conscious AI workflows.

Key Features

• PDF-to-Podcast Workflow
  A four-step pipeline that:

  1. Extracts and preprocesses text from PDFs

  2. Uses LLaMA models to generate podcast-style scripts

  3. Refines the dialogue into natural, conversational tone

  4. Converts scripts into audio using open-source TTS tools like Parler-TTS or Bark

• LlamaCloud Integration
  Seamlessly connects with LlamaCloud to manage model inference and document indexing.

• Modular CLI Toolchain
  Comes with command-line tools and Jupyter notebooks for text extraction, indexing, and backend orchestration—easy to customize and extend.

• Streamlit Frontend
  Provides a user-friendly web UI built with Streamlit, allowing users to upload documents, initiate sessions, and listen to the generated podcast content.

• Observability & Tracing
  Integrated with Jaeger for performance monitoring and interactive session tracing.

Technical Architecture

Notebook Llama follows a two-layer architecture:

• Core Processing Layer

  • Scripts handle text extraction and semantic indexing from documents

  • LLaMA-based models (e.g., LLaMA 3.2, 3.1) generate conversational scripts with simulated multi-speaker dialogue

  • Open-source TTS engines synthesize high-quality audio

• Service & Interface Layer

  • Python-based backend server built with uv, following MCP-style tool orchestration

  • Streamlit frontend for user interactions and playback

  • PostgreSQL and Jaeger run via Docker Compose for session state and observability

• Deployment Environment

  • LLaMA models run on LlamaCloud

  • Uses ffmpeg for audio processing

  • Fully deployable locally or in cloud environments

Project Info & Setup

• GitHub Repo: run-llama/notebookllama

Use Cases

• Document Audio Exploration
  Interact with your PDFs, research papers, or manuals in conversational audio—backed by semantic indexing and retrieval.

• Podcast Content Generation
  Automatically convert PDFs or plain text into multi-voice, AI-narrated podcast episodes.

• Educational Demos & Research
  Perfect for learning or teaching LLM, NLP, TTS, and agent orchestration pipelines.

• Local Deployment & Privacy First
  Run everything locally with your own models and TTS tools—ideal for compliance-sensitive or offline applications.