RAG Frameworks · LearningCircuit

local-deep-research

Local Deep Research is an open-source AI research assistant that runs entirely on local hardware or cloud LLMs, supporting 10+ search engines (arXiv, PubMed, web, private documents) with encrypted storage. It achieves ~95% accuracy on SimpleQA benchmarks using models like Qwen 3.6-27B on consumer GPUs and provides agentic research capabilities with proper citations.

Talk to DEV.co Discuss Implementation

Source: GitHub — github.com/LearningCircuit/local-deep-research

8.7k

GitHub stars

768

Forks

Python

Primary language

MIT

License (OSI-approved)

Key facts

Objective fields from the source. Values we can't verify are shown as “Unknown” rather than guessed.

Field	Value
Repository	LearningCircuit/local-deep-research
Owner	LearningCircuit
Primary language	Python
License	MIT — OSI-approved
Stars	8.7k
Forks	768
Open issues	281
Latest release	v1.8.1 (2026-07-02)
Last updated	2026-07-08
Source	https://github.com/LearningCircuit/local-deep-research

What local-deep-research is

Python-based RAG system using LangGraph for autonomous agent research workflows, compatible with llama.cpp, Ollama, and OpenAI-compatible endpoints. Features SQLCipher-encrypted database, multi-engine search orchestration, document indexing/embedding, and web UI. Deployed via Docker, Docker Compose, or pip; requires AVX-capable CPU (x86-64 or ARM64).

Quickstart

Get the local-deep-research source

Clone the repository and explore it locally.

terminalbash

1git clone https://github.com/LearningCircuit/local-deep-research.git2cd local-deep-research3# follow the project's README for install & configuration

Need it deployed, integrated, or customized instead? DEV.co ships production installs.

Best use cases

Academic Research & Literature Synthesis

Organizations conducting research across arXiv, PubMed, and Semantic Scholar can leverage agentic search to autonomously discover papers, synthesize findings, and generate reports with citations—all locally and privately.

Private Knowledge Base with Q&A

Teams handling sensitive documents (legal, proprietary, healthcare) can build encrypted, searchable libraries indexed locally, then query them alongside public sources without exposing data to external APIs.

Offline Research & Regulatory Compliance

Organizations in air-gapped environments or with strict data residency requirements can perform complex research using local LLMs and cached sources, maintaining full encryption and auditability.

Implementation considerations

GPU memory: 27B models require ~16–24 GB VRAM; quantization or smaller models (7–13B) for 8–12 GB cards. CPU-only inference is significantly slower.
Search engine setup: SearXNG or other search backend must be running separately; plan for configuration, rate limiting, and indexing latency.
SQLCipher encryption: Pre-built wheels included; if compilation fails, fallback to unencrypted SQLite available via bootstrap flag.
Network / Docker: `--network host` on Linux only; Windows/Mac require Docker Compose or explicit port mapping; plan for inter-container communication.
Scaling research agents: LangGraph orchestration can be CPU-bound during multi-step reasoning; monitor LLM and search latency for large result sets.

When to avoid it — and what to weigh

Real-time Web Intelligence Needed — If you need live market data, social media monitoring, or continuously updated news feeds, local-deep-research's search engines may have staleness or limited coverage for fast-moving events.
Multi-Tenant SaaS Without Isolation — Deploying a commercial SaaS service for many users on shared hardware will face resource contention and shared-state security concerns; designed for single-user or small-team local deployments.
Minimal Hardware / Legacy Infrastructure — Requires AVX-capable CPU (2011+), sufficient GPU/CPU for local LLM inference (3090 recommended for 27B models), and Docker/pip tooling; not suitable for embedded systems or pre-2011 CPUs.
Turnkey, Zero-Config Solution — Setup requires orchestrating Ollama, SearXNG, and LDR itself; involves environment variables, API keys, and optional SQLCipher compilation. Not a single-click appliance for non-technical users.

License & commercial use

MIT License (OSI-compliant permissive license). Allows use, modification, and distribution with minimal restrictions; requires preservation of original copyright notice and license text.

MIT permits commercial use without explicit permission, but verify: (1) all dependencies' licenses are compatible with your commercial model (many are permissive, but review transitive deps), (2) use of external APIs (Google, Anthropic, Brave, etc.) may have separate commercial terms, and (3) SQLCipher and optional proprietary LLM services may have additional licensing. Consult legal team for enterprise deployment.

DEV.co evaluation signals

Editorial assessment — not user reviews. Directional, with an explicit confidence level.

Signal	Assessment
Maintenance	Active
Documentation	Adequate
License clarity	Clear
Deployment complexity	Moderate
DEV.co fit	Strong
Assessment confidence	High

Security considerations

Project runs multiple automated scanners (OpenSSF Scorecard, CodeQL, Semgrep, DevSkim, Bearer, OSV-Scanner) and maintains CI/CD checks. Database encrypted via SQLCipher. However: (1) security posture depends on correct deployment (network isolation, API key rotation), (2) local LLM and search backends must be secured separately, (3) no mention of rate limiting, CSRF, or authentication in README (verify web UI security), and (4) private document ingestion assumes trusted input. No penetration test or formal security audit referenced.

Alternatives to consider

Perplexity / SearchGPT

Cloud-based research with real-time web search and GPT-4 class models; no privacy, closed APIs, subscription-based. Use if real-time accuracy and ease-of-use outweigh data residency.

Jan.ai / Ollama Standalone

Lighter local LLM runners; no built-in search, RAG, or research orchestration. Use if you only need LLM inference and will build search/RAG separately.

Langchain + Vector Store (DIY)

Framework for building custom RAG pipelines; requires more engineering to match LDR's research agent, benchmarking, and turnkey UI. Use if you need complete control over architecture.

Software development agency

Build on local-deep-research with DEV.co software developers

Start with Docker Compose (all platforms) or pip install. Supports Ollama, custom LLMs, and 10+ search engines. Full source code, benchmarks, and community support via Discord.

Talk to DEV.co

Related open-source tools

Surfaced by semantic similarity across the DEV.co open-source index.

DeepTutor RAG-Driven-Generative-AI ai-engineering-hub

Related on DEV.co

Explore the category and the services that help you build with it.

RAG Frameworks AI Application Development API Development (Search & LLM Integration)DevOps & Container Orchestration Talk to DEV.co

local-deep-research FAQ

Can I use Local Deep Research on Windows?

Yes: via Docker Desktop (with port mapping workarounds documented in FAQ), Docker Compose, or pip install. PDF export requires Weasyprint/Pango setup; standard SQL export always works.

What GPU is required?

RTX 3090 (24GB) easily runs Qwen 3.6-27B; smaller models (7–13B) fit on 8–12GB cards. CPU-only mode is slow but functional. AVX CPU required (Intel Sandy Bridge / AMD Bulldozer 2011+).

Is my data encrypted?

Database uses SQLCipher (AES-256) by default. Downloaded sources stored locally. External search queries may be logged by search engines (use SearXNG self-hosted to avoid). No TLS in default web UI (use reverse proxy for production).

How accurate is the agentic research?

~95% on SimpleQA (n=500) and 77% on xbench-DeepSearch (n=100) with Qwen 3.6-27B locally. Accuracy varies by model size, search coverage, and research strategy; verify on your use case.

Work with a software development agency

Need help beyond evaluating local-deep-research? DEV.co is a software development agency offering software development services and web development for teams of every size. Our software developers and web developers build custom software, web applications, APIs, and rag frameworks integrations — and maintain them long-term.

Ready to Deploy Private, Agentic Research?

Start with Docker Compose (all platforms) or pip install. Supports Ollama, custom LLMs, and 10+ search engines. Full source code, benchmarks, and community support via Discord.

Talk to DEV.co Schedule a Consultation