ratchet-research

Deep research capability for the Ratchet framework.

Install

pip install ratchet-research

What it does

Give your agent a research question. It decomposes it into sub-queries, searches the web, fetches page content, synthesizes everything into a cited report, and stores the result for future retrieval.

from ratchet.research import ResearchModule

# Register with your agent
agent.register(ResearchModule())

# Or use the pipeline directly
from ratchet.research import research
report = research("What are the latest improvements to transformer attention mechanisms?")
print(report.synthesis.summary)

Pipeline

1. Plan — LLM decomposes question into 3-6 focused sub-queries
2. Search — Web search via Brave, Serper, or DuckDuckGo (auto-detect)
3. Fetch — Extract text content from result pages
4. Synthesize — LLM combines sources into a cited report with confidence rating
5. Store — Persist to JSONL with TF-IDF vector search for retrieval

Search providers

Set one of these environment variables to use a paid search API:

• BRAVE_API_KEY — Brave Search (recommended)
• SERPER_API_KEY — Serper (Google results)

No key? Falls back to DuckDuckGo HTML scraping automatically.

License

MIT