User repository on GitHub
Jev-like decisions. Strong C++ types.
- Forks
- 1
- Last pushed
- Updated Sep 2026
Repo stats from the GitHub API, cached Oct 2026.Homepage
CommunityJev Repos & SDKs
JevT++
Independent C++20 library for typed, model-backed routing, classification and scoring. Compile-time enum schemas define Choice, Noul, Score and Probability answers with abstention; local Laya inference runs through ONNX Runtime or ggml, and an opt-in remote backend speaks the same typed API with hosted inference.
- Stars
- 5
- Language
- C++
- License
- MIT
- Published by
- Community
- Last verified
- Oct 2026
About this project
Application context in. Typed decisions out. A C++20 library for typed model-backed routing, classification and scoring. JevT++ turns runtime text, JSON or application objects into enums, boolean decisions and scores your code can use directly. Define the available answers and their meaning once; supply new context on every call. Optional Laya backends run inference inside your process with ONNX Runtime or native ggml. An opt-in remote backend uses the same typed API with hosted inference.
Quoted from the project's README. Read the full README.
- Capabilities
- Open sourceSelf-hostedRuns locally
Highlights
- constexpr schema and decision_model describe the output vocabulary; inputs arrive at runtime.
- Optional Laya ONNX or ggml inference runs in-process; the hosted backend is opt-in via JEVT_ENABLE_REMOTE.
- Paired CPU benchmark: 139/151ms p50/p95 for one field and 565/605ms for four fields.
- Scripted test backends exercise application behavior without model downloads or paid API calls.
Watch out
MIT licensed. It bundles a Laya adapter rather than the proprietary Jev model and is not an official TypeSafe SDK, so quality depends on the chosen Laya weights; the released v0.2.0 adapter has no GPU path, and the Mario demo did not clear the full game.
Questions about this entry
Short answers drawn from the linked project or page.
No. The core library has no Python runtime, JSON library or ONNX dependency; the Laya inference adapter is optional and enabled at build time, and JSON syntax validation is left to your serializer. CMake exports a relocatable jevt::jevt target for consumers.
A paired i7-10700K CPU benchmark with identical inputs, weights and ONNX Runtime versions: JevT++ at 139/151ms p50/p95 for one field and 565/605ms for four, versus 138/145ms and 584/631ms for the paired Python implementation, showing similar inference latency.
Four: choice returns an enum with a full distribution and entropy confidence; noul returns true, false or abstain with P(true); score returns a modal rubric level plus a fractional expected position; probability returns P(true) for a proposition. Questions are independent, so one answer is not fed into another.
Yes. The input is owned once and shared across the questions in a single Laya batch, and the same binding can be reused as new context arrives on every call. Custom backends can implement predict_batch(), whose default implementation calls predict() sequentially.
The optional remote module uses the same typed API with hosted inference and is enabled at build time via JEVT_ENABLE_REMOTE. By default inference is local, running Laya in-process through ONNX Runtime or native ggml, with CPU or CUDA selection on current main.
choose_async() is a std::async convenience, and the optional batching backend adds bounded workers, microbatching and owned future submission. The Boost.Asio adapter supports completion tokens and co_await; request deadlines and stop tokens are checked by the queue and transports, but cannot interrupt arbitrary running local inference.
No. It is an independent MIT-licensed C++20 library, not the proprietary Jev model or an official TypeSafe SDK. The bundled adapter runs Laya, so model quality depends on the chosen weights; the released v0.2.0 adapter has no GPU path and the Mario demo did not clear the full game.
More Jev resources like this
21GitHub stars
Self-hosted Python SDK and System One-compatible API for runtime-defined Choice, Boolean and Score decisions on pinned open models, across Torch, vLLM, MLX, llama.cpp and WebGPU. Each result records probabilities, token usage, timing, model revision and prompt version, with committed row-level benchmarks.
Jev Repos & SDKs#community#sdk#self-hosted
Community
chengyongru/fastjevPython
chengyongru/fastjevPython1.3kGitHub stars
A local daemon that pulls open decision models and serves them on TypeSafe's /v1/systemone, so a Jev client switches hosts with TYPESAFE_BASE_URL.
Jev Models & Reimplementations#community#rust#open-models
Community
ollaya-dev/ollayaRust
ollaya-dev/ollayaRust882GitHub stars
A Node and TypeScript package that runs Convai's Laya checkpoint through ONNX Runtime, with the same system_one shape as the Python reference.
Jev Repos & SDKs#community#typescript#open-models
Community
receptron/layaTypeScript
receptron/layaTypeScript5GitHub stars
An independent, Jev-inspired System One engine written in C# and .NET 10: a pinned Laya/mmBERT checkpoint runs through pure C# CPU and CUDA kernels with Native AOT, returning typed Choice, Score and Boolean decisions to .NET applications. Probabilities are explicitly uncalibrated.
Jev Models & Reimplementations#community#dotnet#inference
Community
IoTSharp/SezikaC#
IoTSharp/SezikaC#1GitHub stars
Dependency-free Go 1.24+ client for Jev Noul, Choice, and Score questions: ChoiceOf[T] and ScoreOf[L] return the caller's own types and reject labels the question never offered, with validation before sending, retries that honor Retry-After, OpenRouter support, and eleven tested examples.
Jev Repos & SDKs#community#go#sdk
Community
HomayoonAlimohammadi/jev-sdk-goGo
HomayoonAlimohammadi/jev-sdk-goGoSupersonic Labs' 144.3M-parameter decision model on mmBERT-small: choice, score, and noul answers from supplied options in one pass, 73.15% on 2,000 typed decisions, Apache-2.0.
Jev Models & Reimplementations#community#open-models#local
Community
From the Jev guides
Original write-ups on what this entry is for.
Closed weights have not stopped the ecosystem from rebuilding the shape in the open. API-compatible servers, retrained heads on small models, and inference-time tricks that skip generation entirely.
Jev Repos & SDKs#open-models#research#local
Read article
From the Jev community
Posts from builders shipping with Jev right now.