Your PDFs, now a podcast: the open-source NotebookLM rival

What it does Podcastfy is a Python toolkit that ingests multimodal content—websites, PDFs, YouTube videos, images, or even a raw topic—and synthesizes scripted audio conversations via GenAI. It outputs everything from two-minute shorts to longform episodes, using external LLMs to write transcripts and text-to-speech services to render them. The pipeline is exposed as a Python package and CLI, with a FastAPI wrapper currently in beta.

The interesting bit The project’s main differentiator from NotebookLM is automation and control: it is built to be scripted, customized, and scaled rather than clicked through a web UI. It supports over a hundred LLM backends for transcript generation, including local HuggingFace models, and can ground episodes in real-time web search when starting from a topic rather than a document.

Key highlights

• Generates multilingual audio from text, PDFs, images, websites, or YouTube URLs.
• Supports both short clips (2–5 minutes) and longform podcasts (30+ minutes).
• Offers 100+ LLM options for transcript generation, plus local LLM support for privacy.
• Integrates with multiple TTS providers: OpenAI, Google, ElevenLabs, and Microsoft Edge.
• Includes a beta FastAPI server for containerized deployment.

Caveats

• The FastAPI interface is explicitly marked beta and the README documentation for it is sparse.

Verdict Content creators, educators, and developers building automated media pipelines should look here; if you just want a one-off summary and already live inside Google’s ecosystem, NotebookLM is probably less friction.