Browser Use adds Jev routing to its agent, finding flights in 7 seconds
Browser Use put a small decision model inside its agent to handle routing and scoring, finding flights in 7 seconds for $0.0039, with a 10-step guide to building a task router.
browser use put a tiny decision model inside an agent and found flights in 7 seconds for $0.0039. llm creates the work → jev decides what happens next → code executes. most of your agent's expensive calls are just routing, scoring and yes/no checks. which worker goes next, is this source relevant, is the draft ready. they never needed generation. jev reads the state of your system, picks from options you define and returns a typed answer with a probability. $0.042 per million input tokens, zero for output. the receipts are already in. 1,018 research papers classified for $0.08. 500 emails triaged for 3.5 cents. one builder cut a claude session from nearly 1m tokens to 86k in a second. the guide below goes from zero to a working task router in 10 steps: one api key, python 3.12, a json queue on your machine. most builders will keep paying frontier prices for every route and score. the few who split thinking from deciding will ship faster agents for a fraction of the cost. save the framework below
This post quotes another post: “Long read: how to build a Jev-powered agent stack”