Project Files
README
SearXNG Search Plugin for LM Studio
This plugin enables LM Studio assistants to search the web using a local SearXNG instance, providing privacy-focused metasearch capabilities similar to how Vane (Perplexica) utilizes SearXNG.
Prerequisites
-
Running SearXNG Instance: You must have SearXNG installed and running locally (e.g. at
http://localhost:8081to prevent possible conflicts with other services) -
JSON API Enabled: Ensure your SearXNG
settings.ymlincludes:search: formats: - html - json
Installation
- Download this plugin from LM Studio Hub or install manually
- Configure the plugin settings in LM Studio:
- SearXNG URL: Your instance URL (default:Β http://localhost:8081) (if a different port is used, update "config.ts" accordingly. Note that current port setting is "8081" to prevent possible conflicts with other services)
- Default Results Count: How many results to return per query (1-20)
- Timeout: Request timeout in milliseconds
Usage
Once installed, the plugin provides two tools:
search_web: Search the internet using SearXNGfetch_page_content: Retrieve and extract text from specific URLs
The assistant will automatically invoke these tools when web search is needed.
Project Structure
lms-plugin-searxng/ βββ manifest.json βββ package.json βββ tsconfig.json βββ README.md βββ src/ βββ index.ts βββ config.ts βββ toolsProvider.ts
Troubleshooting
- "JSON format not enabled": Ensure your SearXNG settings.yml includes
jsonin search formats - Connection timeouts: Verify SearXNG is running and accessible at the configured URL
- No results: Check that SearXNG has search engines enabled in its configuration
Comparison with Existing Solutions
Unlike the DuckDuckGo plugin which relies on DuckDuckGo's HTML scraping, this SearXNG implementation offers:
- Privacy control: Data never leaves your infrastructure (when using local SearXNG)
- Customizable sources: SearXNG aggregates 70+ search engines configurable by the user
- No rate limits: Local instances aren't subject to external API quotas
- Metadata richness: Access to result scores and source engine attribution
License
MIT