local-deep-research
~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.
A self-hosted AI research assistant that searches the web and academic databases like arXiv and PubMed, then writes cited reports, all on your own computer with your data staying private.
Local Deep Research is a self-hosted AI research assistant that runs on your own computer and keeps your data private. You give it a question, and it searches across the web, academic paper databases like arXiv and PubMed, and your own saved documents, then writes a report with citations. The whole process happens on hardware you control, so nothing is sent to a third-party service unless you choose to connect one.
The tool supports a wide range of AI models. You can run it entirely offline using models loaded through Ollama or llama.cpp, or connect it to cloud providers like Google or Anthropic. The README highlights that using a specific open-weight model on a single consumer graphics card achieved roughly 95% accuracy on a standardized factual question benchmark, placing it among the top-performing research tools on consumer hardware.
Research strategies are configurable. The most capable mode uses an autonomous agent that decides on its own what to search, which databases to query, and when it has gathered enough to write a synthesis. Other modes are available for faster or simpler lookups. Once a research session completes, you can save the sources into a local encrypted library, where they get indexed and become searchable for future sessions. Your document collection grows over time and feeds back into future research.
Installation options include Docker, Docker Compose, and a pip package. The web interface runs at localhost:5000. All stored data is encrypted using SQLCipher, a variant of SQLite with built-in encryption, so the library on your disk is protected at rest.
This is aimed at researchers, students, and anyone who wants a private alternative to cloud-based AI research tools, particularly those working with sensitive documents or academic literature. The full README is longer than what was shown.
Where it fits
- Ask a research question and get a sourced written report drawn from web searches, arXiv, and PubMed, without sending data to any cloud service.
- Build a private growing document library that gets indexed and searched automatically in future research sessions.
- Run fully offline research using a local open-weight model through Ollama on a consumer GPU, with no API keys required.
- Use the autonomous agent mode to let the tool decide what to search, which databases to query, and when it has enough to write a synthesis.