MODSetter/SurfSense: An open source, privacy focused alternative to NotebookLM for teams with no data limit's. Join our Discord: https://discord.gg/ejRNvftDp9

An open source, privacy focused alternative to NotebookLM for teams with no data limit's. Join our Discord: https://discord.gg/ejRNvftDp9 - MODSetter/SurfSense

Summary

SurfSense is an open-source Python desktop app that turns local documents into cited answers and a range of finished artifacts (briefings, slide decks, reports, podcasts, mind maps, and more) without uploading anything. It is positioned as a privacy-focused, air-gapped alternative to Google NotebookLM and is licensed under Apache-2.0.

What it is useful for

SurfSense is useful when engineers, researchers, or analysts need to query and repurpose documents that cannot leave their machine, such as case papers, interview transcripts, internal specs, lecture notes, and unpublished research. The local index (SQLite under ~/.surfsense) and on-device parsing, chunking, and embedding let you ask questions across many files and get answers that cite the source, then export the same material as a .pptx deck, .docx report, .xlsx spreadsheet, PDF, self-contained HTML page, Markmap mind map, flashcards, quiz, or a two-host podcast voiced offline by Kokoro-82M.

It also fits a privacy-sensitive RAG workflow. Outbound connections are off by default with an egress panel to whitelist destinations, there is no telemetry, no crash reporting, and no account to create. The README notes the project tests ingestion with networking disabled. You can point SurfSense at any OpenAI-compatible API or use a bundled local model (Qwen3 in six sizes starting at roughly 0.5 GB), and any API key you provide is stored encrypted with the secret in the OS keychain.

How engineers can use it

The supported path is the desktop app rather than the repo. Download the installer for your platform (Windows .exe, macOS Apple Silicon .dmg, Linux AppImage or .deb) from the GitHub releases page, run it, pick a model, and drop in PDFs, Office files, or images. No Docker, terminal, GPU, or compose file is required. The AppImage auto-updates; the .deb does not.

For self-hosting or integration work, the repo still ships a Docker stack (surfsense_backend, surfsense_web, compose files) and an MCP server in surfsense_mcp, but the README explicitly marks these as community-supported with no SLA. Because the model API is OpenAI-compatible, you can point SurfSense at another local stack such as Jan, LM Studio, Open WebUI, or AnythingLLM and reuse their endpoints.

Documented limitations: video overviews are not built yet, and the hosted web app is being retired (export window closes 18 October 2026). Full install, model, and Studio format details are in the project docs.