Launching today
GateKeep — Decision-Gate Routing Kit
Stop paying frontier prices for yes/no questions
0 followers
Stop paying frontier prices for yes/no questions
0 followers
Cloudflare's Clef and Strands' Decider 2B: small decision models returning structured choices in milliseconds. GateKeep wires that pattern into your stack — a FastAPI decision gate routes every agent tool call. Confident decisions take the cheap path; uncertain ones escalate to the frontier model. Over-budget or kill-switched traffic fails closed (hard 402, kill switch, unknown keys blocked). Kit: gating middleware, 3 routing policies, calibration runbook, sourced price tables.
Built this after the same-day news: Cloudflare shipped Clef/Clef-flash and Strands shipped Decider 2B — small models that return structured decisions in milliseconds. Meanwhile Vercel's gateway data shows open-weight models running 56% of tokens at 14% of spend. The takeaway: stop calling a frontier model to answer yes/no questions. GateKeep is the routing kit I wish existed — gate middleware, policies, calibration runbook, sourced price tables. Happy to answer questions on threshold calibration or the fail-closed design. — Haku