jev-model-router
Route Claude Code model tiers and reasoning effort with one parallel Jev classification.
What Jev decides
Which fast, balanced, or deep tier fits the work, how much reasoning it needs, and whether production risk requires escalation.
The mod asks Jev for a task tier, reasoning score, and production-risk probability, then applies asymmetric confidence gates before changing a subagent model or the main loop’s effort.
- Jev role
- Classify task complexity, required reasoning effort, and operational risk before a Claude Code turn runs.
- Input
- The submitted prompt or subagent task, configured tier criteria, supported effort choices, and routing policy.
- Output
- A tier Choice, effort Score, and risk Noul probability with available confidence values.
- Decision frequency
- At prompt submission for the main turn and when a routed subagent is spawned.
- Fallback
- Keep the engine’s original model and effort on timeout, malformed output, or low confidence; use the built-in classifier when no external key is configured.
How it works
Overview
jev-model-router is an early-access mod inside the claude-code-templates repository. At prompt submission it evaluates three questions in parallel: a Choice over work tiers, a Score for reasoning effort, and a Noul-style risk check. Local policy maps those answers to configured models and supported effort levels.
Downgrades require a higher confidence than upgrades because undersizing a task is costlier than spending extra model capacity. Risk above the configured threshold selects the deep tier. Main-loop model switching is disabled by default because it can invalidate prompt caching, while subagent model routing and main effort routing are enabled. The mod requires Claude Code function hooks and documents that the API may change.
- Language
- TypeScript
- License
- MIT
- Author
- claude-code-templates contributors
Sources
Related projects
Distill
Route coding-agent work, reasoning effort, and context retention through bounded Jev decisions.
What Jev decides
Which configured model and effort should handle a call, what evidence deserves attention, whether another attempt is useful, and what context should be retained.
Jev Review
A code-review pipeline assembled from small typed judgments instead of one long reviewer prompt.
What Jev decides
What is risky, which files and evidence matter, what could fail, how severe it is, and who should review it.
typesafe-computer-use
A native macOS computer-use loop built from local OCR, accessibility data, and bounded Jev actions.
What Jev decides
Which computer action, visible item, site, or off-screen accessibility control to use next.