Local Deep Research — Privacy-First AI Research Agent with 10+ Search Engines

Introduction

Local Deep Research is a self-hosted AI research agent that performs multi-step deep research queries using local or cloud LLMs. It searches across academic databases (arXiv, PubMed), the open web, and your private document collections, synthesizing findings into structured reports. All processing and data storage happens locally with encryption.

What Local Deep Research Does

• Performs iterative deep research across multiple search engines
• Integrates with arXiv, PubMed, Brave Search, SearXNG, and more
• Searches your private document collections alongside public sources
• Synthesizes findings into structured research reports with citations
• Supports both local models (Ollama, llama.cpp) and cloud LLM providers

Architecture Overview

The application is a Python-based research agent with a web UI served via Flask. The research loop works iteratively: it formulates search queries, retrieves results from configured search backends, extracts relevant information, identifies knowledge gaps, and repeats until the research question is adequately answered. A retrieval-augmented generation pipeline handles private document search using local embeddings.

Self-Hosting & Configuration

• Install Python dependencies and configure at least one LLM provider
• Set up search engine API keys or connect to a local SearXNG instance
• Optionally point the private document searcher at your local file directories
• All data is stored locally with optional encryption at rest
• Deploy via Docker for a containerized setup with persistent storage

Key Features

• Multi-source research combining academic, web, and private document search
• Local-first architecture with all data encrypted on disk
• Support for 10+ search engine backends
• Compatible with local LLMs via Ollama for fully offline operation
• Web UI for interactive research sessions with history

Comparison with Similar Tools

• Perplexity — cloud AI search with no local option or private document search
• Tavily — API-focused web search for agents, not a full research pipeline
• Storm (Stanford) — academic research synthesis without private document integration
• GPT Researcher — similar deep research but cloud-dependent, less privacy focus
• Khoj — personal AI assistant with search, different research methodology

FAQ

Q: Can it run fully offline? A: Yes, by using a local LLM through Ollama and a self-hosted SearXNG instance for web search. Private document search works offline by default.

Q: Which academic databases does it search? A: arXiv and PubMed are supported natively, with additional sources configurable through the search engine plugin system.

Q: How does private document search work? A: It indexes your local files using embedding models and performs RAG-based retrieval alongside web search results.

Q: What LLM providers are supported? A: Ollama, llama.cpp, OpenAI, Anthropic, Google, and Mistral are supported.

Sources

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