Local Deep Research
A self-hosted AI research assistant that runs agentic, multi-step research across many LLMs and more than ten search engines, including private documents, with proper citations.
Local Deep Research is an AI research assistant for deep, agentic research, run and controlled locally for privacy. It performs multi-step research using a choice of local or cloud LLMs, such as llama.cpp, Ollama, or hosted providers, together with more than ten search engines, including arXiv, PubMed, and a user's own private documents, and returns answers with citations.
A fully local configuration on a single RTX 3090 running Qwen3.6-27B is reported to reach about 95% on the SimpleQA benchmark and 77% on xbench-DeepSearch. Data is stored locally and encrypted with SQLCipher by default, and a project builds its own searchable knowledge base over time rather than depending on a hosted service.
It installs through Docker, Docker Compose, or pip, with a web UI served locally, and needs an AVX-capable x86-64 CPU (Intel Sandy Bridge, AMD Bulldozer 2011, or newer) or equivalent. It is written in Python and released under the MIT license.