Local Deep Research — Deep Research with Local LLMs and 10+ Search Engines

Introduction

Local Deep Research is an open-source tool that performs multi-source deep research using local or cloud LLMs. It searches across arXiv, PubMed, the web, and your private documents, then synthesizes findings into a grounded summary. Everything runs locally and encrypted.

What Local Deep Research Does

• Performs multi-step research across 10+ search engines and data sources
• Supports local LLMs via Ollama, llama.cpp, and cloud providers
• Searches arXiv, PubMed, Brave, SearXNG, and private document collections
• Synthesizes findings into structured, citation-backed reports
• Keeps all data and processing local with encryption support

Architecture Overview

The system uses an iterative research loop. Given a query, it first generates search sub-queries, dispatches them to configured search engines in parallel, collects and deduplicates results, then uses an LLM to analyze and synthesize findings. If the initial results are insufficient, it generates follow-up queries and repeats the cycle. The final report includes citations with links to source material. All LLM inference can run locally through Ollama or llama.cpp.

Self-Hosting & Configuration

• Install via pip: pip install local-deep-research
• Configure your LLM backend (Ollama, OpenAI, Google, Anthropic) in settings
• Enable search engines by providing API keys or connecting to SearXNG
• Add private document directories for local corpus search
• All configuration stored in a local YAML file

Key Features

• Multi-engine parallel search across academic, web, and private sources
• Supports fully local operation with Ollama or llama.cpp
• Citation-backed reports with links to original sources
• Iterative research loop that refines queries based on initial findings
• Encrypted local storage for research history and document indexes

Comparison with Similar Tools

• Perplexica — open-source AI search engine; Local Deep Research supports more specialized academic sources
• ChatGPT Deep Research — cloud-based; Local Deep Research runs entirely self-hosted
• Khoj — personal AI assistant with search; Local Deep Research focuses specifically on multi-source research synthesis
• Storm (Stanford) — research article generation; Local Deep Research is more focused on quick, grounded answers

FAQ

Q: Does it work completely offline? A: LLM inference can be fully offline with local models. Web and academic searches require internet access by nature.

Q: Which local models work best? A: Models with strong instruction following in the 7B-27B range work well. The project documentation lists tested configurations.

Q: Can I add my own document collection as a source? A: Yes. Point it to a directory of PDFs, Markdown, or text files and it will index and search them alongside web sources.

Q: Is it suitable for academic research? A: It is a research assistant tool. Always verify findings against primary sources before citing in academic work.

Sources

• https://github.com/LearningCircuit/local-deep-research