Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up

jeff (logan-markewich)

Self-hosted drop-in replacement for Jev's System One API, powered by the 400M-parameter GLiFormer model.

LibrarybetaMIT
View on GitHub ↗

Overview

This jeff is a self-hosted implementation of TypeSafe's Jev System One API backed by GLiFormer, a 400M-parameter model from Knowledgator. It supports choice, score and noul questions, and the official typesafe-sdk works against it by setting TYPESAFE_BASE_URL. It is for teams who want Jev-shaped decisions without a cloud dependency or per-call fees.

It runs with uv and Python 3.12, picks CUDA, MPS or CPU automatically, protects the endpoint with configurable API keys, and includes deployment scripts for Modal on L4 GPUs. The README is upfront that it is cheaper to self-host but less accurate than Jev on reasoning-heavy tasks, and links benchmarks. It is MIT licensed. Benchmarks comparing it with hosted Jev are linked from the README so users can judge the trade-off for their own tasks.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this about billing?yes · no
ChoiceWhat is the tone?calm · angry
ScoreHow urgent is this?low · medium · high

How it works

  1. Download GLiFormer weights
  2. Start the server with JEFF_API_KEYS
  3. Point the official SDK or curl at /v1/systemone
  4. Receive Jev-format answers

Strengths

  • Works with the official TypeSafe SDK unchanged
  • No per-call cost once deployed
  • Small model runs on CPU, MPS or modest GPUs

Caveats

  • Less accurate than Jev on reasoning-heavy tasks, per the README
  • Requires managing your own deployment

Getting started

`uv sync --extra dev`, `uv run hf download knowledgator/gliformer-large-v1 --local-dir models/gliformer-large-v1`, then `JEFF_API_KEYS=devkey uv run jeff`.

At a glance

Type
Library
Status
beta
Language
Python
Runtime
HTTP server (local or Modal)
Has code
yes
Has tests
yes
License
MIT

Use cases and integrations

How curated lists describe it

  • Self-hosted runtimes: self-hosted drop-in replacement for TypeSafe Jev powered by GliFormer, exposing native Choice, Score, and Noul decision endpoints without cloud API dependencies.

    yibie/awesome-jev

  • A self-hosted drop-in replacement for TypeSafe.

    wh000wh000/awesome-jev-live

  • A self-hosted drop-in replacement for TypeSafe's jev, powered by GliFormer.

    hellogumbo/awesome-jev

  • Self-hosted implementation of Jev's System One API on the 400M-parameter GLiFormer model. The official SDK works after changing the base URL.

    MrJev/awesome-jev

Frequently asked questions

Is it compatible with the official SDK?

Yes, set TYPESAFE_BASE_URL to the jeff server.

How accurate is it compared to Jev?

The README says it is less accurate on reasoning-heavy tasks and links benchmarks.

What hardware does it need?

It selects CUDA, then MPS, then CPU; Modal deployment uses an L4.

Related projects

  • decidealot

    Docker service exposing local Laya, Von, a CLM head or hosted Jev through one TypeSafe-compatible HTTP and MCP endpoint.

    MCP serverbetaPython
  • edgejev

    Converts open Jev reproductions (Laya, Kev, NanoJev, PlayJev) to INT8 ONNX for offline CPU decisions at about 15.6 ms per question.

    ServicebetaPython
  • go-system-one

    Native Go runtime serving Gemma 4 12B as a Jev-like decision model with hand-tuned PTX kernels and no CGo or CUDA toolkit.

    ProjectbetaGo
  • gojev

    Go harness for System One models: hosted Jev, open Kev in-process via llama.cpp, and Cloudflare Clef, with typed helpers and a CLI.

    ServicebetaGo
  • hearim

    Go gateway exposing the Jev /v1/systemone contract over ordinary LLM backends by reading single-token label logprobs.

    ServicebetaGo
  • intern-decision-mlx

    Runs Intern-Decision-0.8B on Apple Silicon behind a local /v1/systemone endpoint that accepts screenshots for agent decisions.

    ServicebetaPython