SearXNG-lite

Lightweight multi-engine aggregated web search. No Docker, no server, no SearXNG instance required — just a single Python script that queries search engines directly.

26 engines across 9 categories, concurrent execution, JSON + compact text output.

What makes this different

Unlike other SearXNG skills that need a running SearXNG server, this skill:

• Zero infrastructure — no Docker, no SearXNG instance, no HTTP server
• Single file — one Python script (~850 lines) does everything
• Direct queries — sends requests to search engines in-process via httpx
• Hot-reload config — edit config.yml to toggle categories, changes apply instantly
• Concurrent — queries multiple engines in parallel (up to 5 threads)
• Deduplication — merges results from multiple engines, scores by cross-engine overlap

Requirements

• Python 3.10+
• httpx — HTTP client (pip3 install httpx)
• lxml — HTML parser (pip3 install lxml, pre-installed on macOS)
• (Optional) socksio — for SOCKS proxy support (pip3 install socksio)
• (Optional) pyyaml — for config parsing (pip3 install pyyaml; falls back to built-in parser)

Quick start

# Install dependencies
pip3 install httpx lxml

# Search
python3 scripts/search.py "your query"

# List all engines
python3 scripts/search.py --list

How to search

python3 scripts/search.py "query"                          # default (general + knowledge engines)
python3 scripts/search.py "query" -c dev                   # by category
python3 scripts/search.py "query" -c dev,academic          # multiple categories
python3 scripts/search.py "query" -e github,arxiv          # specific engines
python3 scripts/search.py "query" --all                    # all enabled engines
python3 scripts/search.py "query" -l zh-CN                 # Chinese results
python3 scripts/search.py "query" -n 5                     # limit results
python3 scripts/search.py "query" --compact                # title + url + snippet text output
python3 scripts/search.py --list                           # show all engines & categories

All paths are relative to this skill's directory.

Arguments

Without -e or -c, searches general + knowledge categories.

Configuration

Edit config.yml in the skill directory to customize behavior:

# Proxy for engines that need it (Google, YouTube, Reddit, etc.)
# Supports: http, https, socks5, socks5h
# Leave empty to disable — proxy-required engines will fail silently.
proxy: "socks5h://127.0.0.1:1080"

# Category toggles
categories:
  general: true       # bing, brave, duckduckgo, google🌐, startpage, yahoo
  knowledge: true     # wikipedia, wikidata, wolframalpha🌐
  dev: true           # github, gitlab, stackoverflow, hackernews, reddit🌐, huggingface🌐, mdn
  academic: true      # arxiv, semantic_scholar, google_scholar🌐, crossref
  news: false         # bing_news, reuters
  video: false        # youtube🌐
  images: false       # unsplash
  social: false       # lemmy🌐
  translate: false    # lingva🌐

Proxy setup

Some engines (marked with 🌐) require a proxy to access. Three ways to configure:

1. config.yml (recommended): set proxy: "your-proxy-url"
2. Environment variable: set HTTPS_PROXY=your-proxy-url
3. CLI flag: pass --proxy your-proxy-url per search

Priority: CLI flag > config.yml > environment variable.

If no proxy is configured, 🌐-marked engines will fail silently and results from other engines are still returned.

Config hot-reload

The config file is read on every search call. No restart needed — just edit and save.

If config.yml is missing, the script falls back to a default set of engines: bing, brave, duckduckgo, wikipedia.

Categories and engines

🌐 = requires proxy. Without proxy, these engines are skipped.

Output format

Default JSON:

{
  "query": "search term",
  "results": [
    {"title": "...", "url": "...", "content": "...", "engines": ["bing","brave"], "score": 2}
  ],
  "result_count": 15,
  "elapsed": 2.1,
  "engines_used": ["bing", "brave", "wikipedia"],
  "errors": []
}

--compact outputs human-readable text: numbered title, URL, snippet, engine tags.

score = number of engines that returned this result. Higher = more relevant.

Typical agent workflows

• General research: python3 scripts/search.py "topic" -n 5
• Find a library: python3 scripts/search.py "image processing python" -e github
• Academic papers: python3 scripts/search.py "attention mechanism" -c academic -n 5
• Tech discussions: python3 scripts/search.py "topic" -e hackernews,reddit
• AI models: python3 scripts/search.py "text-to-speech" -e huggingface
• Dev docs: python3 scripts/search.py "fetch API" -e mdn,stackoverflow
• Chinese search: python3 scripts/search.py "大语言模型" -l zh-CN -n 5
• News: python3 scripts/search.py "AI regulation" -c news