Cloudflare Opens WebSearch API Beta, Replacing Agents' URL Guessing with Routed Search
大模型之路 · wechat · 2026-10-13
What happened
Cloudflare launched its WebSearch API in public beta on October 2, shifting agent web access from "guess a URL and fetch" to routed queries.
How it works
- It's a routing layer, not a proprietary index: you submit a query with a provider param; results return in a unified format
- Three launch backends: Ceramic.ai (default), Exa, Linkup
- Runs on AI Gateway — search and inference share logs, quotas, caching and rate limiting; supports BYOK with no markup
- Three call paths: REST endpoint, Worker AI binding (env.AI.websearch()), or via AI Gateway
- Unified items array (url/title/description + metadata) means switching backends is a one-string change
Three accounts to run
- Cost: per 1k requests — Ceramic $0.25, Linkup $5, Exa $7; agents can fire a dozen queries per task
- Compliance: zero data retention is per-provider — Ceramic and Linkup support it, Exa doesn't
- Reliability: no SLA in beta and no cross-provider benchmarks; test all three with small real traffic first
The bigger play
All three backends must follow Cloudflare's "verified crawler" standard and return source links. Cloudflare blocks unlicensed AI scraping on one side and sells compliant retrieval to agent builders on the other. The author's takeaway: agent search is becoming a swappable gateway component — what stays locked in is your policies, logs, and audit trail, not the search backend.
More from coding & agent
- Daniel Jeffries: only ~2% can actually use AI agents, mass adoption is 10-20 years away — Dan_Jeffries1 · 2026-10-13
- Fully local AI short film mimics Wes Anderson style with a detailed open pipeline — Mist3r-Robot · 2026-10-13
- Controlling his Mac's terminal from a phone via t3 code beta while waiting at speech therapy — NERDDISCO · 2026-10-13
- LLM cache verifier drift: reject-only calibration tripled error rate; 5% audits fix it — Reasonable_Royal_621 · 2026-10-13
- A style guide for making Claude-like models write normally, and its linguistic traps — gleech · 2026-10-13
- Open-source desktop agent dodaios runs local Qwen models via ollama — and shows where 16GB models fail — InternationalGap3698 · 2026-10-13