Show HN: Search for AI agents with usefulness score built in

Category: ai-ml

Tags: ai-agents, web-search, ranking, token-efficiency, sales-automation

Score: 5.3/10 (Innovation: 6, Technical: 4, Documentation: 7, Utility: 4)

Xtract is a middleware layer that scores web search results for usefulness before an AI agent reads them, aiming to reduce token costs and prevent stale facts from reaching outputs. It combines embedding similarity, a small model check, search rank, citation history, and freshness detection into a single weighted score, with a simulated per-page pricing mechanism and a reuse loop where scores are shared across agents.

Target audience: AI agent developers, especially teams building sales research or SDR agents

Repository: https://github.com/Ayobamiu/usefulness-search · Python · 3 stars

View on Hacker News