JevCodeResearch & reimplementation
Open reimplementations and experiments on the System One decision model itself.
Projects in this category
- Mapika/deciderGitHub ↗★ 1.2kPythonA family of System One-style models fine-tuned from Qwen3.5, designed for one-pass typed decisions with calibrated probabilities.
- wfzyx/vonGitHub ↗★ 923PythonThe open-source System One decision model. Sub-15ms, non-autoregressive, local drop-in alternative to TypeSafe Jev.
- YannQi/R-4BGitHub ↗★ 186—A 4B open decision model (Reflex-4b): benchmarked at 2-3x faster than Jev with roughly 5% lower accuracy, Apache 2.0.
- kshetrajna12/reflexGitHub ↗★ 174PythonA small open decision model: state + typed questions → calibrated probabilities, re-created on Qwen3.5.
- ikermoel/open-alternative-jevGitHub ↗★ 74PythonOpen-source alternative to TypeSafe's Jev: a System One style model layer that gives typed, calibrated decisions from any open-weights LLM in one forward pass (HF + vLLM), with honest benchmarks
- Mapika/decider-2bGitHub ↗★ 60—A 2B open decision model (Decider-2b): benchmarked at ~10x faster than Jev, fully local, roughly 5% lower accuracy.
- r-ms/mini-jevGitHub ↗★ 60Pythonmini-Jev: what a Jev-style typed-decision interface looks like on a frozen Qwen3-4B — read the option letter's logits instead of generating JSON. Preregistered experiment, results, teaching bench.
- zhengxuyu/litjevGitHub ↗★ 47PythonTurn any off-the-shelf LLM into a Jev -like decision layer
- zhihz/openjevGitHub ↗★ 37PythonLocal bilingual probability decisions from context, questions, and candidate answers. Independent research preview inspired by TypeSafe Jev.
- GiesN/typesafe-jev-workflowGitHub ↗★ 11PythonAn async LangGraph example that classifies mocked emails as invoice-related or general.
- us/jev-localGitHub ↗★ 9PythonLocal Jev-compatible evaluation server: POST /v1/systemone with typed noul/choice/score, open weights, no waitlist
- scienthoon/jev-ood-calibrationGitHub ↗★ 6PythonIndependent calibration test of TypeSafe's Jev on a task it cannot have seen: 900 rule-generated support tickets (choice / score / boolean) plus 3 public benchmarks via Vercel AI Gateway. Raw responses, ECE with noise floor, temperature refit, per-type sign of miscalibration. Reproducible for ~$0.06.
- jgridifier/jev-research-evalGitHub ↗★ 3HTMLReproducible Jev Ultrafast research-browser eval harness + field note (QC’d cases, suite runner, report generator). Not investment advice.
- ably-labs/jev-pongGitHub ↗★ 1TypeScriptPong where the ball moves one step per model decision. Jev vs LLMs via Vercel AI Gateway, every player and agent on an Ably channel.
- miounet11/openjevGitHub ↗★ 0—Can we run something like Jev on a 3090 at home?
- yodablocks/jev-orderby-benchGitHub ↗★ 0PythonDoes ORDER BY over a Jev probability put rows in a defensible order? Independent ranking, calibration and invariant measurements of TypeSafe AI's Jev: passes six pre-registered gates on 360 labeled rows, fails four of six on graded product relevance.
Indexed in the Jev community ecosystem directory, data snapshot 2026-11-02.
Frequently asked questions
What is Research & reimplementation?
"Research & reimplementation" is a project category in the Jev ecosystem with 16 projects listed. Open reimplementations and experiments on the System One decision model itself.
Which projects in "Research & reimplementation" should I look at first?
Sorted by stars, start with: Mapika/decider, wfzyx/von, YannQi/R-4B, kshetrajna12/reflex, ikermoel/open-alternative-jev. Each has its own page covering positioning, language, license and alternatives in the same category.
Other categories
Keep going on this site
Try, docs, cases, and pricing link to each other. The next page is on this site.
- Site directoryThe full list of the pages above.
- EcosystemCommunity projects by category.
- Jev coverageEach entry names the decision and its evidence.
- Daily casesThe next batch of coverage cases, each day.
- CommunityPublic posts, each with its source.
- BuildsNotes on model and site updates.
- PrimitivesChoice, Noul, and Score together.
- PatternsRouting, fan-out, and composite scoring.
- SDKsHTTP, Python, and JavaScript.
Use clavue-jev
- Try a judgmentSend a state and read the answer.
- Live scenesTetris, six rapid judgments, and a sort. Copy the call and run it again.
- Judgment LabSupport triage, release review, compliance routing — edit and run on the spot.
- Compareclavue-jev and jev-1.13.0 answer the same question.
- DemosJudgment examples you can edit.
- PricingInput is $0.042 per million tokens. Output is free.
- APICall POST /v1/judge with a key.
Docs
- IntroductionWhat this model is for.
- Quick startThe smallest call that runs.
- System OneWritten here: a state plus one typed question. The model is clavue-jev.
- StateThe text you send in.
- ConfidenceHow sure the answer is, and when to act on it.
- ChoicePick one mutually exclusive option.
- NoulA yes/no question, returned as the probability of yes.
- ScoreRate along a scale.
- Intent routingOne Choice call dispatches the request.
- Confidence routingAct when confidence is high. Hand off when it is low.
- Fan-outAsk several questions of one state.
- Composite scoringFold several scores into one ranking.
- HTTP APIThe request shape without an SDK.
- Python SDKCall a judgment from Python.
- JavaScript SDKCall a judgment from JavaScript.
Cases
- Use-case mapWhich primitive people use for which job.
- GuardrailsA yes/no check before an action runs.
- Function callingPick which function to call.
- Classify with confidenceDo not act when the class is unsure.
- Semantic findFind the passages that match.
- Parallel questionsAsk the whole set in one call.
- Hierarchical classificationCoarse class first, then the fine one.
- Citation checkWhether a sentence is supported by the source.
- Entity alignmentWhether two names are the same thing.
- Skill suggestionChoose which skill fits the moment.