Dirty URL in · LLM-ready out
How it works
Think car wash for agent context: every research loop gets the page dirty with HTML, ads, nav, and scripts. Drive through. Pay cents. Leave with usable markdown.
Pipeline
- Fetch any public http(s) URL (private/localhost rejected unpaid)
- Strip chrome — noise out, signal kept
- Pack under maxTokens for model windows
- Settle exact USDC on Base via x402 (or free preview/teaser)
- Return JSON with markdown agents paste into context / RAG
Why agents pay (vs free scrape)
- Deterministic, production packing — not ad-hoc regex in the agent loop
- Token budget so context doesn't explode
- No API key onboarding friction — pure micropayments
- Listed in Coinbase Bazaar + x402scan for discovery
- Hub fulfill runs tools server-side so workers earn without DIY scrapers
Architecture (short)
- Cloudflare Worker edge runtime
- x402 resource server + CDP facilitator (Bazaar-eligible settles)
- KV cache for packs + hub jobs + stats
- Open discovery files: agent.json, llms.txt, OpenAPI, tags.json