JEV-APP-PC
Local Anthropic-compatible proxy where Jev scores each Claude Code turn and simple work is routed to cheap or free models.
Overview
JEV-APP-PC contains jev-router, a local Anthropic-compatible gateway for Claude Code or any Messages-API client that uses Jev as a System One classifier to keep simple work off the Claude quota. Its design record explains that on a Claude subscription, switching between Claude models saves nothing because all models share the same quota and switching discards the prompt cache, so the goal is to spend Claude only on work that needs it.
On each new human turn Jev answers one ordinal Score question giving a distribution over simple, standard and structural. Simple tasks such as questions, explanations, single-file edits and renames go to a cheap OpenAI-compatible provider (Ollama Cloud's free plan by default, or Groq or OpenRouter); standard feature work without risk can go to a larger cheap model; and structural work such as architecture refactors, heavy test design, concurrency, security and performance goes to Anthropic unchanged. The cheap route requires confidence, and anything it cannot carry falls back to Anthropic.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Score | How complex is this turn's work? | simple · standard · structural |
How it works
- Claude Code sends requests to the local router
- On a new human turn, Jev scores complexity
- Confident simple or standard work goes to a cheap provider (translated to chat completions)
- Structural or uncertain work goes to Anthropic byte-for-byte
Strengths
- Saves Claude quota for work that needs it
- Free-plan default for cheap route
- Clear written rationale (ADR) for the design
Caveats
- Cheap models may produce lower-quality results
- Requires managing multiple provider keys
At a glance
- Type
- Proxy
- Status
- prototype
- Language
- TypeScript
- Runtime
- local HTTP proxy
- Has code
- yes
Use cases and integrations
How curated lists describe it
Local Anthropic-compatible proxy that routes Claude Code prompts to free models…
wh000wh000/awesome-jev-live
Frequently asked questions
What does Jev decide?
Whether each new turn is simple, standard or structural work.
Why not route between Claude models?
On a subscription they share one quota and switching discards the cache, so it saves nothing.
Where does simple work go?
To a cheap OpenAI-compatible provider, Ollama Cloud by default.
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