Jevonian
Local OpenAI/Anthropic-compatible proxy where one Jev call picks the model route and thinking level for each coding-agent turn.
Overview
Jevonian is a local model router and protocol gateway that sits between a coding agent (Claude Code, Cursor, Codex) and your providers. It exposes one endpoint at 127.0.0.1:8787; requests to jevonian/auto are routed per turn to a plan, execute or utility/chat route with a suitable model, while explicit routes and pinned models remain available. Providers can be API keys or subscriptions such as Claude Pro/Max and ChatGPT Codex, with live quota meters.
Deterministic code first filters candidates by wire protocol, context window, thinking-level floor and exhausted quota windows. Then a single Jev call answers both the route and the thinking level from session state: recent messages and tool results, consecutive errors, context headroom, quota health, candidate capabilities and cache-switch penalties. A minConfidence setting marks low-confidence routes in the local ledger rather than silently accepting them, and the ledger records the serving model, reason, tokens and estimated cost. Pinned models and routing mode off skip Jev entirely. Status is v0.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which route (plan, execute, utility/chat) and model should serve this turn? | plan · execute · utility/chat |
| Choice | Which thinking level should this turn use? | — |
How it works
- Agent sends a request to jevonian/auto
- Code filters candidate models by protocol, context, thinking floor and quota
- One Jev call picks route and thinking level from session state
- Low-confidence routes are marked in the ledger
- Request is forwarded to the chosen provider; ledger records model, tokens, cost and reason
Strengths
- Keep your existing agent; one local endpoint for many providers
- Quota- and cache-aware routing
- Deterministic filtering before the model is asked
- Local ledger for cost and reasoning
Caveats
- v0 status
- Routing quality depends on Jev's read of session state
- Requires a Jev routing key in addition to provider credentials
Getting started
npm install --global jevonian && jevonian, then open http://127.0.0.1:8787, add a provider and routing brain, and point your agent at jevonian/auto.At a glance
- Type
- Proxy
- Status
- beta
- Language
- Go
- Runtime
- local proxy service with web UI
- Has code
- yes
Use cases and integrations
How curated lists describe it
Coding agents: local OpenAI/Anthropic-compatible proxy where one Jev call picks both the model route and the thinking level for `jevonian/auto` from session state, quota health, candidate capabilities, and cache-switch penalties, after deterministic code has filtered candidates and while pinned models, explicit `jevonian/<route>` requests, and `routing.mode: "off"` skip Jev entirely; `minConfidence` marks a low-confidence route in the ledger rather than accepting it, and the ledger records the serving model and why.
yibie/awesome-jevLocal OpenAI / Anthropic / Responses-compatible proxy where one Jev call answers both the model route and the thinking level for `jevonian/auto`, from session state (recent messages and tool results, consecutive errors, context headroom, quota, candidate capabilities, cache-switch penalties); deterministic code filters candidates and owns every threshold first, a pinned model or explicit `jevonian/<route>` skips Jev entirely, and each decision is recorded with the serving model, reason, token usage, and estimated cost.
AnotiaWang/awesome-decision-modelsLocal OpenAI/Anthropic-compatible proxy where one Jev call answers model route and thinking level for `jevonian/auto`, after deterministic code has already filtered candidates by wire protocol, context window, thinking-level floor, and spent quota windows; `minConfidence` marks a low-confidence route in the ledger instead of silently accepting it, and Jev is skipped entirely for pinned models, explicit `jevonian/<route>` requests, and `routing.mode: "off"`.
cobanov/awesome-jevLocal OpenAI-compatible proxy that sits between a coding agent and its providers and asks Jev, per turn, which routed model and thinking level should serve it.
hellogumbo/awesome-jevSits between a coding agent and its providers, picks a model for each turn against quota and cache economics, and records the decision locally.
MrJev/awesome-jev
Frequently asked questions
What does Jev decide?
For each turn, which route and model should serve it and at what thinking level.
Can I bypass routing?
Yes. Pinned models, explicit jevonian/<route> requests and routing mode off skip Jev.
Is it an agent?
No. It is a router and gateway; your agent owns the workflow and your provider runs the model.
Related projects
Cost-, cache- and quota-aware LLM router that classifies each turn with a Jev-class model, prices every route, and checks cheap answers.
ProjectprototypePythonLocal-first, outcome-aware model router for coding harnesses that picks a capability tier per task with Jev and learns from history.
ProjectprototypeRust- 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 TypeScript/Bun launcher that asks Jev which model and reasoning effort each Codex, Claude Code or Pi turn should use.
ProjectbetaTypeScriptGo CLI model router that uses one Jev analysis to recommend a model, harness and effort level, then selects deterministically.
CLIbetaGoClaude Code and Codex plugin where Jev judges each turn for model routing, delegation, evidence search and compaction.
ProjectbetaTypeScript