LocalJev
GitHub Next's Bun server that exposes a Jev-compatible /v1/systemone API by asking DiffusionGemma for probabilities via a chat endpoint.
Overview
LocalJev, from GitHub Next, is a local Jev-compatible POST /v1/systemone API written in TypeScript for Bun, backed by DiffusionGemma through an OpenAI-compatible chat completions endpoint such as oMLX. It exists because OpenJev's exact logit-reading approach depends on unmerged vLLM request extensions that ordinary servers do not expose.
LocalJev takes a portable route: it translates the state and typed Jev questions into a classification prompt, asks DiffusionGemma for a JSON probability scalar or vector, validates the result and retries malformed output, normalises vectors, and computes Jev-compatible choices, expected scores and entropy-based confidence before returning the normal Jev response shape. The README is explicit that this is wire-compatible but not mathematically equivalent to a logit read, because the probabilities are generated or self-reported by the model, and advises evaluating calibration on your own workload before relying on it for consequential decisions. A readiness endpoint checks the configured model is available.
What the decision model decides
| Type | Question | Answers |
|---|---|---|
| Choice | Which team should handle this? | billing · technical · sales |
| Score | How frustrated does the customer appear? | Calm · Frustrated but civil · Very angry |
How it works
- Run an OpenAI-compatible server with DiffusionGemma (e.g. oMLX)
- Start LocalJev with Bun
- POST Jev-format requests to /v1/systemone
- Model returns JSON probabilities; LocalJev validates and normalises
- Jev-shaped response is returned
Strengths
- Works with standard chat endpoints
- Jev wire compatibility
- Validation and retry for malformed output
Caveats
- Probabilities are self-reported, not logit-derived
- Calibration must be checked per workload
- Requires a local DiffusionGemma server
Getting started
bun install, cp .env.example .env (set the upstream key), bun run start, then POST to http://127.0.0.1:8080/v1/systemone.At a glance
- Type
- Service
- Status
- prototype
- Language
- TypeScript
- Runtime
- Bun server
- Has code
- yes
Use cases and integrations
How curated lists describe it
Local Jev-compatible `/v1/systemone` server for Bun that asks DiffusionGemma for probabilities, from GitHub Next.
MrJev/awesome-jev
Frequently asked questions
How does it get probabilities?
By asking DiffusionGemma to output a JSON probability vector, then validating it.
Is it equivalent to OpenJev?
No, it is wire-compatible but probabilities are self-reported rather than read from logits.
Who made it?
GitHub Next.
Related projects
Docker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.
MCP serverbetaPythonConverts open Jev reproductions (Laya, Kev, NanoJev, PlayJev) to INT8 ONNX for offline CPU decisions at about 15.6 ms per question.
ServicebetaPythonNative Go runtime serving Gemma 4 12B as a Jev-like decision model with hand-tuned PTX kernels and no CGo or CUDA toolkit.
ProjectbetaGoGo harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.
ServicebetaGoGo gateway exposing the Jev /v1/systemone contract over ordinary LLM backends by reading single-token label logprobs.
ServicebetaGoRuns Intern-Decision-0.8B on Apple Silicon behind a local /v1/systemone endpoint that accepts screenshots for agent decisions.
ServicebetaPython