GitHub radar
LocalJev: Run Jev Typed Classification Locally for Free
GitHub Next built a local server with the Jev protocol: send a message and several typed questions at once, get all answers in one call — running on your Mac through DiffusionGemma, no API fees.
LocalJev is a local server implementing the TypeSafe Jev protocol, made by GitHub Next. Jev is an API that takes a message and a list of typed questions — which team should handle this, how urgent is the issue, how frustrated is the customer — and returns answers to all of them in one call. LocalJev translates the questions into a classification prompt, sends it to a local DiffusionGemma model through an oMLX server, and returns the result in the same format as the paid TypeSafe Jev API. The main difference: the original Jev reads probabilities directly from model logits, while LocalJev has the model generate them as text, so results are approximate. The authors note that calibration should be verified on your own data before using this for consequential decisions.
Why a vibe-coder should care
If your app or agent needs to classify incoming messages — routing support tickets, detecting urgency, scoring sentiment — LocalJev lets you do this on your own Mac without paying per API call. It works well for straightforward, repetitive classification where occasional errors are acceptable. For high-stakes decisions, check calibration on your own data first.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Deploy localjev from https://github.com/githubnext/localjev following the README — set up Bun, run an oMLX server with DiffusionGemma 26B-4bit, start localjev, and show me a test request classifying a support message with three questions: department, urgency, mood. Ask me if you need API keys.
Requires a Mac with 32 GB RAM for DiffusionGemma 26B 4-bit. Does not run on a 16 GB machine.
Open on GitHub▌ More finds