An OpenAI- and Anthropic-compatible endpoint. A fast edge model reads every request and routes it: answer instantly at the edge, run a web search at the edge, or escalate to a frontier model — so you never pay a frontier model to say hello.
Chit-chat, short facts, formatting — answered on Workers AI in a blink. Nothing upstream is touched.
llama-3.1-8bCurrent-info questions run a search tool at the edge and come back synthesized + sourced — still no frontier call.
edge + web_searchOnly genuinely hard reasoning escalates to a real upstream model, with an edge fallback if it's down.
deepseek-chatcurl https://oneprompt.site/v1/chat/completions \
-H "authorization: Bearer $ONEPROMPT_KEY" \
-H "content-type: application/json" \
-d '{"model":"oneprompt-auto","stream":true,
"messages":[{"role":"user","content":"What changed in Cloudflare Workers this week?"}]}'