LearningCircuit/local-deep-research

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

What it solves

Local Deep Research (LDR) provides a private, self-hosted alternative to AI research assistants. It allows users to perform deep, agentic research across the web, academic papers, and personal documents without sending sensitive data to centralized AI providers, while ensuring all findings are backed by proper citations.

How it works

LDR uses a combination of LLMs (via Ollama or OpenAI-compatible endpoints) and various search engines (such as SearXNG, arXiv, and PubMed) to automate the research process. It offers two primary strategies:

  • Pipeline modes: Used for fast fact-finding and quick summaries.
  • LangGraph Agent Strategy: An autonomous agent that decides which specialized search engines to use and when to synthesize information, allowing for more complex analysis and higher accuracy.

Users can also build a local knowledge base by downloading sources into an encrypted library, which LDR then indexes and embeds for future retrieval-augmented generation (RAG) during research sessions.

Who it’s for

  • Privacy-conscious researchers who need to perform deep analysis without telemetry or tracking.
  • Academic and technical professionals who require high-accuracy results with citations from reputable sources.
  • Developers who want to integrate agentic research capabilities into their own applications via a REST API.

Highlights

  • Fully Local Execution: Can run entirely on a single RTX 3090 using local LLMs.
  • Encrypted Data: Uses SQLCipher with AES-256 encryption for per-user databases and API keys.
  • Agentic Research: Adaptive search strategies that switch between engines based on findings.
  • Comprehensive Source Integration: Supports academic engines (arXiv, PubMed), general web search (SearXNG), and local document indexing.
  • Journal Quality System: Automatically scores journal reputation and detects predatory journals using OpenAlex and DOAJ.
  • Knowledge Base Building: Ability to download and index sources to create a compounding, searchable personal library.

Related

  • Project
    ruc-datalab/DeepAnalyze

    An agentic LLM designed for autonomous data science, capable of automating the entire pipeline from data preparation to the generation of professional research reports.

  • Project
    amaiya/onprem

    A privacy-focused Python toolkit for document intelligence that enables local-by-default LLM execution for sensitive data, with optional cloud integration.

  • Project
    u14app/deep-research

    An AI-powered research tool that automates the gathering of information from the web and local files to generate comprehensive, in-depth reports.

  • Project
    yzr278892/arxiv-daily-researcher

    An LLM-powered research monitoring system that automatically collects, filters, and analyzes papers from ArXiv and other sources to generate personalized daily reports.

  • Project
    trycompai/crm

    An open-source, agentic-first CRM where AI agents autonomously research leads, record factual evidence, and manage follow-ups instead of relying on manual human entry.