Jev Model Router
A Claude Code module that uses Jev to route model tier, reasoning effort, and risky-action judgment for agent tasks.
Added to Jevfast
How it uses Jev
A Claude Code hook classifies a submitted coding task before its turn, asking Jev for a model-tier Choice, a reasoning-effort Score, and a risky-action Noul. It maps the typed answers and confidence to model/effort settings and separately routes spawned agents. Provider errors keep the original engine request.
What Jev decides
Coding task routingExample answers · not a recorded Jev response · Source ↗
Question 1 · tier
YOUR APP
INSTRUCTION
Choose the cheapest model tier that can complete this task well.
STATE
Current submitted task text and routing context provided by the Claude Code function hook.
JEV · CHOICE
- fast
- balanced
- deep
Question 2 · effort
YOUR APP
INSTRUCTION
Score how much step-by-step reasoning is needed.
STATE
Current submitted task text and routing context provided by the Claude Code function hook.
JEV · SCORE
0
Question 3 · risky
YOUR APP
INSTRUCTION
Judge whether carrying out the task changes production, moves real money, or alters unrestorable data.
STATE
Current submitted task text and routing context provided by the Claude Code function hook.
JEV · NOUL
YesNo
App workflow
- Receive a task
A Claude Code prompt submission triggers the router hook before the model turn.
- Classify the task
Jev returns typed tier, effort, and risk judgments in one request.
- Apply routing policy
The hook checks confidence and risk overrides, then sets the requested model and effort for the turn.
- Handle later work
Subagent spawn events can be classified separately; provider failure leaves the original request in place.
Why it is interesting
The hook separates cost routing, reasoning effort, and real-world action risk into three typed judgments. Its confidence gates are asymmetric, and a high-risk answer can override the cheaper tier; it also avoids a mid-session main-model switch by default.