~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.
Python Package (pip) Installation Guide
Note: For most users, Docker is preferred as it handles all dependencies automatically. pip install is best suited for developers or users who want to integrate LDR into existing Python projects.
Quick Install
# Step 1: Install the package
pip install local-deep-research
# Step 2: Setup SearXNG for best results
docker pull searxng/searxng
docker run -d -p 8080:8080 --name searxng searxng/searxng
# Step 3: Install Ollama from https://ollama.ai
# Step 4: Download a model
ollama pull gemma3:12b
# Step 5: Pin the SearXNG URL (which also marks it operator-approved —
# private/localhost engine URLs are otherwise blocked by default since
# v1.10.3) and start the web interface. The URL becomes read-only in the
# web UI; docs/SearXNG-Setup.md lists the alternatives, e.g. an origin
# allowlist.
export LDR_SEARCH_ENGINE_WEB_SEARXNG_DEFAULT_PARAMS_INSTANCE_URL=http://localhost:8080
ldr-web
Open http://localhost:5000 after a few seconds.
SearXNG returns no results? If the logs show
SearXNG engine disabled: instance URL … is a private / loopback / link-local address, the allowlist env var above is missing from the environment runningldr-web. See SearXNG-Setup for all approval options.
SQLCipher (Database Encryption)
LDR uses SQLCipher for AES-256 encrypted databases. Pre-built sqlcipher3 wheels are available for Windows, macOS, and Linux — most users won't need to compile anything.
- Full setup instructions: SQLCipher Install Guide
- Development fallback:
LDR_BOOTSTRAP_ALLOW_UNENCRYPTED=truepermits standard SQLite only when SQLCipher is unavailable or nonfunctional. It does not disable working SQLCipher. In the fallback, database contents (including stored API keys) are plaintext and account passwords are not verified. Use it only for isolated local development, not shared or family hosting. See fallback limitations. - Docker: Includes SQLCipher out of the box — no extra setup needed.
Optional Dependencies
MCP Server
For integration with Claude Desktop or Claude Code:
pip install "local-deep-research[mcp]"
Platform Notes
Windows PDF Export: PDF export requires Pango/Cairo system libraries. See the WeasyPrint installation guide for setup instructions.
CJK characters in PDF exports: WeasyPrint resolves glyphs through the host's installed fonts. If your research results contain Chinese, Japanese, or Korean characters and they disappear from the downloaded PDF, install a CJK font package:
- Debian/Ubuntu:
sudo apt install fonts-noto-cjk && fc-cache -fv- Fedora/RHEL:
sudo dnf install google-noto-sans-cjk-fonts && fc-cache -fv- Alpine:
apk add font-noto-cjk- macOS: ships with PingFang / Hiragino — no install needed.
- Windows: ships with Microsoft YaHei / SimSun — no install needed.
Docker users on the official image do not need to do anything;
fonts-noto-cjkis bundled.
Emoji in PDF exports: Emojis in the markdown are rendered through the host's emoji font. If they appear as empty boxes ("tofu") in the PDF, install an emoji font package:
- Debian/Ubuntu:
sudo apt install fonts-noto-color-emoji && fc-cache -fv- Fedora/RHEL 10+:
sudo dnf install google-noto-color-emoji-fonts && fc-cache -fv- RHEL 9 / CentOS Stream 9:
sudo dnf install google-noto-emoji-color-fonts && fc-cache -fv(package was renamed togoogle-noto-color-emoji-fontsin EL10)- Alpine:
apk add font-noto-emoji(the package name omits "color" but ships the colorNotoColorEmoji.ttf)- macOS / Windows: ships with the OS — no install needed.
Docker users on the official image do not need to do anything;
fonts-noto-color-emojiis bundled.
Development from Source
For contributing or running from the latest code, see the Development Guide.