Route each agent task to the cheapest model or tool that fits
Ask Jev one Choice question per task to pick a model, tool or subagent, with code enforcing permissions and fallbacks.
DifficultyAdvanced
Setup time1 day
CategoryDecision models (Jev)
The use case
Teams running coding or support agents pay frontier-model prices for tasks a smaller model could handle. These routers put the candidate models, tools and subagents in one list and ask Jev which fits the next task, while code enforces availability, permissions, risk rules and confirmation. Some only switch at task boundaries and log every choice so routing can be audited. An engineer reviews the logs, compares quality and cost per route each week, and keeps high-risk work pinned to the stronger model.
Example workflow
- List candidate models, tools and subagents with short descriptions
- At each task boundary, ask Jev a Choice question over the list
- Apply permission and risk rules in code before running
- Log the choice and outcome
- Review cost and quality per route weekly
Tools
- Jev (TypeSafe AI)
- OpenRouter
- Claude Code
- Codex CLI
Risks and guardrails
- Routing down and back up can cost more than not routing
- Quality drops may show up late
- Bad candidate descriptions skew choices
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- JevRouter by BillionsBobby on Made with Jev: madewithjev.com/builds/jevrouter
- JevRouter: original post or repo by BillionsBobby: github.com/BillionsBobby/JevRouter
- pi-jev-router (win4r) by Chao Qin on Made with Jev: madewithjev.com/builds/win4r-pi-jev-router
- pi-jev-router (win4r): original post or repo by Chao Qin: github.com/win4r/pi-jev-router
- codex-jev-router by tiandee on Made with Jev: madewithjev.com/builds/codex-jev-router
- codex-jev-router: original post or repo by tiandee: github.com/tiandee/codex-jev-router
- OpenRouter (official page): openrouter.ai (checked 2026-11-09)