rsearch

Self-hosted research agent. Ask a question — get a synthesized answer from the web.

Automatically picks the cheapest strategy that can answer your query, from a quick search to full browser automation.

Strategies

Strategy	Trigger	Pipeline	Cost
search_only	Most queries	SearXNG → LLM	~$0.002
search_then_crawl	"read full article", "detailed"	SearXNG → Crawl4AI → LLM	~$0.005
search_then_browse	"compare prices", "fill form"	SearXNG → Browser-Use → LLM	~$0.05–0.10
browse_direct	URL in query, "go to site"	Browser-Use → LLM	~$0.05–0.15

85%+ of queries go through search_only — fast and nearly free.

Stack

• SearXNG — self-hosted meta-search (Google, Bing, DDG, Wikipedia)
• Crawl4AI — fast headless page extraction
• Browser-Use — browser automation for interactive pages
• Claude — answer synthesis (via Claude Code CLI, Anthropic API, or OpenAI)

Setup

1. Start SearXNG

docker-compose up -d

2. Install dependencies

pip install -r requirements.txt
playwright install chromium

3. Configure

cp .env.example .env

# Provider: claude_code | anthropic | openai
LLM_PROVIDER=claude_code

ANTHROPIC_API_KEY=sk-ant-...   # only if using anthropic
OPENAI_API_KEY=sk-...          # only if using openai

SEARXNG_URL=http://localhost:8080

LLM_CHEAP=claude-haiku-4-5
LLM_SMART=claude-sonnet-4-5

Usage

CLI

python -m agent.main

> what is the USD to EUR rate today
> read the full Wikipedia article about Roman Empire
> go to https://example.com and summarize

MCP server — expose as a tool for Claude Code or any MCP client

python mcp_server.py

~/.claude/mcp.json:

{
  "mcpServers": {
    "rsearch": {
      "command": "python",
      "args": ["/path/to/rsearch/mcp_server.py"]
    }
  }
}

Available tools:

• research(query) — run a query, strategy is picked automatically
• research_strategy(query) — preview which strategy would be used

Requirements

• Python 3.10+
• Docker
• Chromium (playwright install chromium)