Jev in production › Model and agent routing
Skill and CLI that runs a panel of local AI coding CLIs in parallel; opt-in Jev routing sends the task text to Jev to choose which CLI model runs it, using model profiles, evidence and subscription quota.
Alloy routes work to the best qualified model for the job, then verifies the result. The goal is high quality with fewer revisions. Subscription capacity and token cost break ties between suitable workers.
Alloy uses the coding CLIs already authenticated on your machine: Codex, Claude, Grok, Antigravity (agy), Muse Code and Cursor. Opt-in routing chooses one worker for a consult, or jointly selects an independent Maker and Checker for managed execution. Explicit multi-model panels remain available for difficult questions whose competing approaches need comparison.
For the project's own README, linking back here: