Local Deep Research (LDR) is an open-source, AI-powered research assistant designed for deep, agentic research workflows that prioritize privacy, security, and user control. Its primary goal is to enable users to perform sophisticated research tasks—such as answering complex questions, synthesizing information from multiple sources, and building a personal knowledge base—entirely on local hardware, without reliance on external cloud services unless explicitly configured.
LDR stands out for its ability to run advanced large language models (LLMs) locally, such as Qwen3.6-27B, achieving impressive accuracy benchmarks (approximately 95% on the SimpleQA dataset and 77% on xbench-DeepSearch) on consumer-grade GPUs like the RTX 3090. It supports a wide range of LLMs and inference backends, including llama.cpp, Ollama, and cloud-based providers, giving users flexibility in model selection and deployment.
The system integrates with over ten search engines, covering academic sources (arXiv, PubMed, Semantic Scholar), general web (Wikipedia, SearXNG), technical repositories (GitHub, Elasticsearch), news outlets, and premium engines like Google (via SerpAPI), Brave, and Tavily. Users can also index and search their own private documents, making LDR a powerful tool for both public and personal research.
A key feature is the agentic research mode, particularly the LangGraph Agent Strategy. In this mode, the LLM autonomously determines which search engines to query, when to synthesize findings, and how to adapt its strategy based on the information retrieved. This approach enables comprehensive, citation-backed reports that draw from a diverse set of sources. Users can choose from over 20 research strategies, ranging from quick fact-finding to in-depth academic analysis.
LDR emphasizes security and privacy at every level. Each user’s data is stored in an isolated, SQLCipher-encrypted database using AES-256 encryption, ensuring zero-knowledge privacy—even server administrators cannot access user data. The application contains no telemetry, analytics, or tracking, and only initiates network connections for user-driven actions such as search queries or LLM API calls. Supply chain security is reinforced through signed Docker images, SLSA provenance, and SBOMs, with a transparent security review process and extensive use of automated security scanners.
The platform offers a rich set of features for research management and productivity: users can save, search, and revisit past research sessions; export results as PDF or Markdown; subscribe to automated research digests on custom schedules; and leverage real-time updates via WebSockets. Advanced capabilities include LangChain integration for custom vector stores, a REST API with per-user authentication, benchmarking tools, analytics dashboards, and a journal quality system that scores sources for reputation and predatory risk.
Installation is flexible, supporting Docker, Docker Compose (with GPU or CPU options), pip, and Unraid. LDR is cross-platform, running on Linux, Windows, and macOS. It provides both Python and HTTP APIs for programmatic access, as well as command-line tools for benchmarking and rate limit management.
In summary, Local Deep Research is a comprehensive, privacy-first research assistant that empowers users to conduct advanced, multi-source research locally, build and search their own knowledge bases, and maintain full control over their data and workflows.
