jev-router-playground
Browser playground: configure a pool of OpenRouter models, let Jev pick one for a task, then judge whether its pick was right.
Overview
Jev Router Playground is an interactive model-routing experiment. You build a pool of OpenRouter models, adding a routing description for each and seeing its pricing and context window; give Jev a task; and Jev selects the best-fit model with a full probability distribution across candidates. You can run all candidates to compare answers, timing and approximate cost, pick the answer you prefer, and see whether it matches Jev's pick, with each run exportable as JSON.
Across many tasks, this lets you measure how often Jev's routing agrees with your judgment and whether running only its choice preserves answer quality. It is a static site with no build step. Keys live in session storage; the OpenRouter key goes directly to OpenRouter, while the Jev key passes through an included Cloudflare CORS proxy by default, because TypeSafe's API does not return CORS headers on error responses. The proxy can be self-hosted or bypassed.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which candidate model is the best fit for this task? | — |
How it works
- Build a candidate pool with routing descriptions
- Enter a task; Jev picks a model with probabilities
- Optionally run all candidates
- Pick the best answer and record match or mismatch
Strengths
- Lets users check routing against their own judgment
- Exportable observations
- No build step
Caveats
- Default Jev calls pass through a public CORS proxy
- Needs both TypeSafe and OpenRouter keys
- Running all candidates costs more
Getting started
python3 -m http.server 8000, open http://localhost:8000 and add keys in Settings.At a glance
- Type
- App
- Status
- prototype
- Language
- JavaScript
- Runtime
- Static web app
- Has code
- yes
- License
- MIT
Use cases and integrations
How curated lists describe it
A model-routing playground where Jev picks a candidate and users compare the resulting answers.
logicrw/awesome-jev-projects
Frequently asked questions
What does Jev decide?
Which model in your pool best fits the task.
Why is there a proxy?
TypeSafe's API does not return CORS headers on error responses, so browser calls need a proxy.
Can I measure routing accuracy?
Yes. Each run records whether your preferred answer matched Jev's pick.
Related projects
Deterministic evidence selector for noisy Codex investigations; its earlier Jev subagent router is kept for the record, not recommended.
ProjectresearchResearch framework (DART) testing whether Jev's probabilities are a reliable signal to answer, escalate or abstain.
ProjectresearchPythonPython tool that measures where to route decisions between Jev and a larger model on your own data, then routes by confidence.
BenchmarkbetaPythonJev as an LLM router on RouterArena and LLMRouterBench: the router beat the best single model, but a no-Jev ablation matched it.
BenchmarkresearchPythonBenchmark of LLM-native structured output (Qwen on Cerebras) versus Jev on seven application tasks: errors, latency, tokens and cost.
BenchmarkresearchClassmethod DevelopersIO independent test of Jev for LLM model routing, reporting 40 successful calls and 10 of 10 correct classifications.
Articleresearch