Jev-Model-Router-Claude-Code
Claude Code hook and skill (German docs) where Jev picks the cheapest sufficient Claude tier and whether to delegate to a subagent.
Overview
Jev Model Router for Claude Code is companion material to a German video on three Jev use cases in Claude Code. A UserPromptSubmit hook sends the prompt text to Jev, which answers two typed questions: the cheapest sufficient tier (haiku, sonnet, opus or fable) and the probability that the task depends on the conversation so far.
The hook then injects a line telling Claude to DELEGATE to a subagent with the named model, or to handle it in the session. Delegation happens only when the tier is below the session model and the task is self-contained. A /jev skill toggles the router and shows its decisions. Only the prompt text (up to 12,000 characters) is sent to TypeSafe; no files, code or history. The router is fail-open: if Jev does not answer within 5 seconds, the prompt passes through unchanged. Installation is done by pasting a prompt that has Claude copy files and merge hook settings.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which is the cheapest Claude tier that can handle this task? | haiku · sonnet · opus · fable |
| Yes / no | Does the task depend on the conversation so far? | yes · no |
How it works
- Hook sends the prompt text to Jev on submit
- Jev returns a tier and a context-dependence probability
- If the tier is cheaper and the task is self-contained, Claude is told to DELEGATE
- Otherwise Claude handles it in the session
- Fail-open after 5 seconds
Strengths
- Sends only prompt text, no files or history
- Fail-open design never blocks work
- Toggle and inspect decisions with /jev
Caveats
- Documentation in German
- Requires jq, curl and a TypeSafe key
- Companion material for a video rather than a maintained package
Getting started
Clone the repo, start claude in the folder and paste the setup prompt from the README; then enable with /jev on.At a glance
- Type
- Project
- Status
- prototype
- Language
- Bash
- Runtime
- Claude Code hook and skill
- Has code
- yes
Use cases and integrations
How curated lists describe it
Begleitmaterial zum Video „Jev + Claude Code: 3 Use Cases".
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
The cheapest sufficient Claude tier for a prompt and whether the task depends on earlier conversation.
What is sent to TypeSafe?
Only the prompt text, up to 12,000 characters.
What if Jev is slow?
After 5 seconds the router fails open and the prompt runs unchanged.
Related projects
- jev-router★ 593
Per-turn model routing for Claude Code and Codex: wraps the real CLI and lets Jev send simple turns to fast models and hard ones to strong.
ProxybetaJavaScript Claude Code mod that sets model and effort per turn and subagent from a prompt classifier: Jev by default, or Clef, OpenAI or Ollama.
ProxybetaTypeScript/Bun launcher that asks Jev which model and reasoning effort each Codex, Claude Code or Pi turn should use.
ProjectbetaTypeScriptCLI, Claude Code plugin and Pi virtual model that pick the right model per prompt from one Jev call with a complexity floor.
PluginbetaTypeScriptClaude Code mod that picks reasoning effort per prompt with a Haiku judge, optionally Jev, and shows cache and context state.
ProjectbetaTypeScriptGo CLI model router that uses one Jev analysis to recommend a model, harness and effort level, then selects deterministically.
CLIbetaGo