Jev and Herdr: The Best Way to Route Models (Must-Try Workflow)
Nidhi Singh demonstrates using TypeSafe AI's Jev model inside a custom CLI router and Herdr agent runtime to automatically select harnesses, models, and reasoning efforts.
Original by Nidhi SinghAgent workflowsIntermediate17 min 11 secPublished
Jev acts as a fast decision engine using three parallel question output types (choice, score, and boolean) rather than chat-based generation.
A custom CLI router leverages Jev to select coding harnesses, target models, and reasoning budgets without manual switching.
Session-level routing rather than per-message switching is used to preserve prompt caches and manage token consumption across planning and execution phases.
Worth knowing
The custom routing CLI shown is a personal prototype requiring early-access API credentials, and its decisions reflect prompt heuristics rather than standardized router benchmarks.