Small decisions. Interesting possibilities.Submit contentSubmit

How it uses Jev

Jev receives the fresh prompt, current model and context token count, and available model IDs. It scores complexity dimensions and chooses an exact model ID; routing policy applies confidence and context rules before the CLI starts the turn.

What Jev decides

Illustrative Jev decisionExample answers · not a recorded Jev response · Source ↗
Question 1 · model
YOUR APP
INSTRUCTION

Choose the cheapest exact available model that can complete the request in one pass.

STATE

New request text, current model and context tokens, with the default Haiku, Sonnet, and Opus model IDs available.

JEV · CHOICE
  1. claude-haiku-4-5-20251001
  2. claude-sonnet-5
  3. claude-opus-5
Question 2 · task complexity
YOUR APP
INSTRUCTION

Score overall task complexity including ambiguity, scope, and blast radius.

STATE

New request text, current model and context tokens, with the default Haiku, Sonnet, and Opus model IDs available.

JEV · SCORE
09

App workflow

  1. Inspect turn

    Read prompt, current model, context size, and available model tiers.

  2. Ask Jev

    Request complexity Scores and a model Choice in one call.

  3. Apply routing policy

    Use confidence, overrides, context size, and available tiers to settle on a model.

  4. Launch CLI turn

    Pass the selected model through the existing CLI interface.

Why it is interesting

It routes only fresh turns while preserving native CLI behavior, and explicitly accounts for the prompt-cache cost of switching models.