What it does
Local Deep Research (LDR) is a research assistant you run yourself. You ask a complex question and it searches across the web, academic sources such as arXiv and PubMed, and your own documents, then writes a report with citations. Strategies range from quick pipeline modes to an agentic mode in which the model decides what to search and when to synthesize. It works with local model runners (Ollama, LM Studio and llama.cpp) and with cloud providers. Search engines include free ones such as arXiv, PubMed, Semantic Scholar, Wikipedia and SearXNG, and premium ones such as Tavily, Google through SerpAPI and Brave Search. Sources you download go into an encrypted per-user library that is indexed, so later research can combine them with the live web. It installs with Docker, Docker Compose or pip, and the README says it works on Windows, macOS and Linux. Access is through a web UI on localhost, a Python API, an authenticated REST API and an MCP server for tools such as Claude Desktop and Claude Code. The README states there is no telemetry, and it describes an egress setting that is a best-effort guard rail rather than an air gap. It reports about 95% on SimpleQA with a locally run Qwen3.6-27B model, and notes the samples are small.
Capabilities
- Cited reports built from web, academic and personal-document searches
- Quick summary, detailed research, report generation and document analysis modes
- Works with Ollama, LM Studio, llama.cpp or cloud model providers
- Encrypted per-user library that indexes downloaded sources for later searches
- Web UI, Python API, REST API and MCP server
- Research subscriptions with daily, weekly or custom schedules
When to use it
- Literature or topic research where you want cited output and control over where your data goes
- Searching private documents together with the live web
- Letting Claude Desktop or Claude Code call a research tool through MCP
- Comparing local models on deep-research tasks using the project's benchmark dataset
