NotebookLM without the Google lock-in

What it does

InsightsLM is a self-hosted alternative to Google’s NotebookLM. You upload documents, ask questions grounded strictly in those sources via RAG, and generate AI audio summaries—including podcast-style discussions—with citations linked back to the original text. The frontend is a React/TypeScript application, while Supabase handles the database and storage, and N8N workflows orchestrate the backend logic and AI calls.

The interesting bit

The project is essentially sophisticated glue: it stitches together a “vibe-coded” Loveable/React frontend with Supabase edge functions and N8N automation nodes to replicate a closed-source product. The authors openly pitch this as a commercial template, framing RAG chatbots as one of the few reliably monetizable generative-AI patterns for businesses.

Key highlights

• Document chat with verifiable citations jumping to source text
• Podcast and audio summary generation from uploaded materials
• Self-hosted stack using Supabase, N8N, and cloud LLMs (OpenAI/Gemini)
• Separate fully-local fork available using Ollama, Whisper, and CoquiTTS
• MIT-licensed and positioned as a customizable commercial starter kit

Caveats

• Requires cloud AI services by default; local LLM support lives in a separate repository
• Podcast generation needs FFMPEG installed on a self-hosted n8n instance and is explicitly unsupported on n8n Cloud
• Supabase recently changed its authentication system, forcing edge functions to handle JWT verification internally instead of using the platform’s native flag

Verdict

Worth a look if you want a private, customizable research assistant and are comfortable maintaining a Supabase-and-N8N backend. Skip it if you expect a managed, zero-config drop-in replacement for NotebookLM.