This project provides an AI-powered research assistant that automates deep, iterative analysis of complex questions across diverse knowledge sources, including academic databases and private documents. It's designed for researchers, students, and power users seeking comprehensive, cited reports with flexible LLM and search engine integration, prioritizing local execution for privacy.

How It Works

The system performs iterative research cycles, generating follow-up questions to refine understanding. It retrieves full webpage content, not just snippets, and integrates with academic sources like arXiv and PubMed, providing inline citations and source verification. Users can leverage local LLMs via Ollama or cloud-based models (OpenAI, Anthropic), with flexible search capabilities including academic databases, web search (via SearXNG or API keys), and local RAG for private documents.

Quick Start & Requirements

• Installation: pip install local-deep-research and playwright install. Ollama is required for local models.
• Local LLM: Ollama must be installed and a model pulled (e.g., ollama pull gemma3:12b).
• Web Search: SearXNG (via Docker) or API keys for providers like Brave Search or SerpAPI are recommended for non-academic searches.
• Running: ldr-web for the web interface or ldr for the CLI.
• Docs: Docker Usage Guide, Programmatic Access Tutorial

Highlighted Details

• Supports local LLMs (Ollama, vLLM, LMStudio, LlamaCPP) and cloud LLMs (OpenAI, Anthropic).
• Integrates with academic sources (arXiv, PubMed, Semantic Scholar) and general web search.
• Enables local RAG for searching private document collections.
• Offers both a web interface and a command-line interface.

Maintenance & Community

• Active development with a clear migration path from v0.1.0.
• Community support via Discord server and Subreddit.
• Contributions are welcomed via Pull Requests to the dev branch.

Licensing & Compatibility

• Licensed under the MIT License.
• Permits commercial use and linking with closed-source applications.

Limitations & Caveats

• Non-academic web searches require a configured SearXNG instance or search provider API keys.
• Initial setup for local LLMs and search engines may require additional configuration steps.