Deep Research (Luke Swade)
Self-hosted AI research agent with cited reports, a searchable library, and local LLM support through Ollama, LM Studio, llama.cpp, or MLX. MIT licensed.
Deep Research is a self-hosted AI agent for people who want web research with traceable evidence and a library they control. It runs in Docker on your own machine or server, searches the web in successive rounds, and uses gaps in its findings to decide what to investigate next. Reports include citations, per-source notes with quoted evidence, and suggested follow-ups.
Research depth ranges from quick answers based on search snippets to longer investigations that read full pages. The agent follows references to find primary documents, reads YouTube captions and Reddit posts with comments, and merges duplicate articles into a single source. Recency filters help focus a run on a chosen period.
You can use a local LLM through Ollama, LM Studio, llama.cpp, vLLM, or MLX, or connect an OpenAI-compatible cloud API such as DeepSeek or OpenAI. A smaller model can handle source notes while a larger one plans the research and writes the report. Web searches and page retrieval need internet access; local model support doesn't make the research process offline. Cloud models receive the material sent in LLM requests.
The browser interface shows live progress and estimates based on earlier runs. Its library supports keyword and semantic search, and an Ask page answers questions across saved research with citations. Topics can refresh daily, with reports focused on changes. An optional Telegram bot starts research and delivers results.
The project is open source under the MIT license. Research records stay on disk as Markdown files and can be exported as PDF, a self-contained HTML page, or an interactive single-file report.
