LearningCircuit

local-deep-research: Research Assistant for Local and Cloud LLMs

Local Deep Research runs agentic research across web, academic, and private document sources, then produces cited reports. It supports local and cloud LLMs, with encrypted per-user data and configurable controls over network access.

View on GitHub Compare
Stars
9.2k
Forks
839
Stars, last 30 days
+287
Last commit
today
Contributors
89

Our take

Activeregular commits and releases

Health score

67/100

How it is scored
Activity
90
Community
59
Issues
67
Pull requests
52

+287 stars in the last 30 days

  • Releases more often than 93% of the projects we track
  • Merges more pull requests than 92% of the projects we track
  • More contributors than 76% of the projects we track
  • Slower to merge pull requests than 86% of the projects we track

This is an actively maintained, feature-rich research tool with frequent releases, but adoption should account for its concentrated contributor base and the limits of its egress safeguards.

Good fit if

  • You want agentic, citation-oriented research across academic, web, and private sources.
  • You need provider choice, including local inference, and are prepared to configure your search and model endpoints.
  • Your deployment can meet the Python and Node.js requirements, and you value MIT-licensed software with encrypted user databases.

Look elsewhere if

  • You need a guaranteed air gap: private-only and strict egress controls are best-effort safeguards, not a hard network boundary.
  • Your team requires project work to be broadly distributed, since most contributions come from one person.
  • You need a research assistant that does not require configuring an LLM and search backend.
All health signals
Last commit2026-10-10 (today)
Commits, last 90 days100+
Releases, last 12 months100+ (latest v1.10.7, 2026-08-28)
Contributors89 (top contributor: 78% of commits)
Issues closed, last 90 days17+ (typically closed in 3 days)
Pull requests merged, last 90 days88 (typically merged in 4 days)
Project age20 months

Checked on 2026-10-11 with the GitHub API.

Project history

Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.

Selected snapshot

Stars

9,599

Forks

930

Change since Nov 17, 2026: +3 stars·+5 forks

Overview

Local Deep Research is a self-hostable research assistant that searches the web, academic indexes, and private document collections to answer complex questions with citations. It offers both quick research workflows and an agentic strategy that can select search engines as it investigates.

It suits researchers and teams who want to combine their own knowledge base with external sources, while choosing which LLM provider to use. Local models can keep inference on the user’s hardware, but web searches still involve the configured search engines unless runs are restricted to private sources.

Key Features

  • Produces quick summaries, detailed research, and structured reports with citations.
  • Searches sources including arXiv, PubMed, Semantic Scholar, Wikipedia, GitHub, and SearXNG.
  • Connects to private documents, local libraries, and LangChain-compatible retrievers.
  • Supports local LLMs through Ollama, LM Studio, and llama.cpp, as well as cloud providers and compatible endpoints.
  • Stores each user’s data in a separate SQLCipher-encrypted database, with per-user LLM API keys encrypted there too.
  • Provides a web interface, REST API, Python API, and an MCP server for local assistant integrations.
  • Offers configurable egress scopes and flags the limitations of its in-process network controls.

Use Cases

  • Researchers can search academic sources and create cited literature summaries, using local models when their hardware supports them.
  • A team can query its own indexed documents alongside web sources when building an internal knowledge base.
  • Privacy-conscious users can run inference locally and use private-only research settings when they need to avoid public search sources.
  • Developers can call research workflows from Python or the REST API, or expose local research tools to an MCP-compatible assistant.

What you need

Detected in the repository

  • Node.js >=24.0.0 (from package.json)
  • Python >=3.12,<3.15 (from pyproject.toml)
  • A Dockerfile, so it can run in a container; a Compose file sets up the related services
  • A test suite and automated checks on GitHub Actions

License in plain words

MITpermissive

  • Commercial use: yes
  • Modify and redistribute: yes
  • You must keep: the copyright and license notice
  • Share your changes: no

A summary, not legal advice: the LICENSE file is what applies.

Getting Started

With Docker Compose, start the CPU-only setup with:

curl -O https://raw.githubusercontent.com/LearningCircuit/local-deep-research/main/docker-compose.yml && docker compose up -d

Then open http://localhost:5000. See the README for GPU setup, provider configuration, and installation alternatives.

Alternatives

  • SurfSense: SurfSense centers on searching local documents and generating study or work materials, while local-deep-research also investigates web and academic sources.
  • DeepTutor: DeepTutor focuses on tutoring and study workflows, while local-deep-research focuses on multi-source research and cited report generation.
ProjectLanguageLicenseStarsStatus
local-deep-researchPythonMIT9.2kActive
SurfSensePythonApache-2.016.6kActive
DeepTutorPythonApache-2.041.8kActive

Considerations

  • The project is young but active: its recent commit and release activity is strong, and pull requests are typically merged within days. It also has a substantial open-issue backlog, so users should check whether unresolved issues affect their deployment.
  • The project has many contributors, but contributions are concentrated heavily in one person, which is an ownership and continuity consideration.
  • Python 3.12 through 3.14 and Node.js 24 or newer are listed requirements. Docker and Compose are supported; a GPU is optional, but local-model performance depends on available hardware and the model chosen.
  • SQLCipher encryption protects stored per-user data, but it does not make network traffic private by itself. Search queries can reach configured upstream engines, and the project explicitly describes egress controls as best-effort rather than an air gap.
  • Local endpoints can relay prompts to hosted models without LDR detecting that relay, and a self-hosted SearXNG can forward queries to internet search providers.
  • The MIT license permits broad use, subject to its license terms. Teams should review their own obligations and deployment requirements before adopting it.

Found this useful?

Share it with someone who would like local-deep-research.

Comparisons