SearXNG Self-Hosted Search

**Quick judgment**: This skill is a solid, privacy-friendly way to give the assistant web search via a self-hosted SearXNG instance. It’s powerful and flexible but assumes the user can run and maintain SearXNG (typically via Docker), so it is best suited to technically comfortable users or organizations. **What it does well** - Uses a self-hosted SearXNG instance as a meta-search engine, aggregating results from 70+ sources. - Supports categories (general, images, news, videos, IT, science), pagination, and language/region filters. - Returns structured JSON (title, URL, snippet, engines, score, category), which is ideal for tool-driven agents. - Prioritizes privacy by avoiding third-party API keys and central tracking. **Key risks and limitations** - **Setup burden**: Requires SearXNG to be running and correctly configured (Docker, `settings.yml` with JSON format, `SEARXNG_URL`); non-technical users may struggle. - **Content risk**: Like any web search, results can include outdated, biased, or unsafe content; the assistant must not treat results as inherently trustworthy and should apply normal critical filtering. - **Configuration mistakes**: Misconfiguring SearXNG (e.g., exposing it publicly without auth, logging queries, or leaving defaults like `secret_key` unchanged) can weaken privacy or security. - **Engine rate limits**: Some upstream search engines may rate-limit or fail; users may see incomplete or inconsistent results depending on SearXNG’s configuration and traffic. **Recommended scenarios** - You already run (or are willing to run) SearXNG locally or on a server and want the assistant to have web access without using commercial search APIs. - You work in a privacy- or compliance-sensitive context and need all search traffic to stay within your own infrastructure. - You want meta-search capabilities (multiple engines, categories, language filters) exposed cleanly to an AI agent for research or automation workflows.