Private AI for Research Papers with Ollama

With Ollama the language model runs on your own computer or on your local network, so your papers, notes and questions never leave your personal space. That suits unpublished manuscripts, papers under peer review, NDA work and sensitive data, and it costs nothing per question. Personal data is more valuable than ever, and this keeps yours private.

Setting it up in practice

  • Install Ollama and download a model with ollama pull.
  • In Research Assistant, open Settings → AI, choose Ollama and enter the model name. Leave the address empty for Ollama’s default.
  • Any OpenAI-compatible server on localhost, such as LM Studio or llama.cpp, works the same way.
  • For Discover, pick a model with tool support.
Settings → AI with Ollama selected as the provider
Settings → AI with Ollama selected as the provider

What runs on your computer

  • Chat and summaries, through Ollama.
  • Search by meaning, with an embedding model bundled with the app.
  • Text recognition with a specialized OCR module that makes scanned papers searchable.
  • Full-text search, notes and highlights, stored as plain files.

Local models usually have a short context, so the app sends the passages that match your question rather than the whole paper. Answers still cite their pages and quotes are still checked. If your model allows a longer context, raise Whole-paper chat in Settings → AI.

What we still need to fetch online

  • Metadata lookup on import sends a DOI or title to Crossref, OpenAlex and similar catalogues.
  • Reference lists, retraction checks and the optional daily feed query the same catalogues.

None of these send the text of your papers, notes or chats. The app has no telemetry or usage tracking.

Try it on your own library

Free for personal use and students, 14.99€ once for work. Windows, macOS and Linux.

Free Download