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

AnyDecisionModel

Swift package for typed decisions (probability, choice, score) over local MLX or Core AI models, Clef decision heads, or the Jev API.

LibrarybetaApache-2.0
View on GitHub ↗

Overview

AnyDecisionModel is a Swift package for typed decisions: yes/no probabilities, choices among options and scores on ordinal scales. Its API follows the model-and-session pattern of Apple's Foundation Models framework and Hugging Face's AnyLanguageModel. A Choosable enum can define options, and results carry per-option probabilities and confidence.

It has four backends behind one protocol: MLX runs a small language model on Apple silicon and reads each decision from next-token probabilities in one forward pass; MLX decision heads run models with a trained head such as Cloudflare's Clef; Core AI does the same on macOS 27 and iOS 27; and the Jev backend calls TypeSafe's /v1/systemone or any server implementing it. It requires Swift 6.2+, with the core and Jev backend also working on Linux.

What the decision model decides

TypeQuestionAnswers
Yes / noDoes this convey urgency?yes · no
ChoiceWhich team?billing · technical · sales
ScoreHow frustrated?Calm · Frustrated · Very angry

How it works

  1. Create a model (MLX, Core AI, Clef head or Jev)
  2. Open a DecisionSession with the state
  3. Call probability, choice or score
  4. Use the typed result in Swift code

Strengths

  • Same API for on-device and hosted decisions
  • Fully offline option on Apple silicon
  • Typed Swift enums for options

Caveats

  • Local small-model accuracy will differ from Jev
  • Requires recent Swift and OS versions

Getting started

Add the AnyDecisionModel Swift package; for Jev set TYPESAFE_AI_API_KEY or TYPESAFE_API_KEY.

At a glance

Type
Library
Status
beta
Language
Swift
Runtime
library (macOS, iOS, Linux)
Has code
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • A Swift package for typed decisions from language models (probabilities, choices, and scores), with support for local MLX models and the TypeSafe Jev API.

    hellogumbo/awesome-jev

  • Swift package with one typed-decision protocol over two backends: Jev's `/v1/systemone`, or a small model running locally on Apple silicon through MLX.

    MrJev/awesome-jev

Frequently asked questions

Can it run without the network?

Yes, the MLX and Core AI backends run locally.

How are local decisions computed?

From next-token probabilities in one forward pass, or from a trained decision head like Clef.

Which key does the Jev backend use?

TYPESAFE_AI_API_KEY or TYPESAFE_API_KEY, or one passed to the initializer.

Related projects

  • jevper

    Python wrapper that answers System One and OpenAI Decisions API questions using any OpenAI-compatible chat model.

    LibrarybetaPython
  • 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
  • jev

    TypeScript library for one-line local decisions with open models (kev-0.6b, kev-4b, open-jev), no API keys or cloud.

    ProjectbetaTypeScript
  • jev-bisect

    Python library that finds numeric answers by bisection, asking Jev higher, lower or exactly at each guess.

    ProjectbetaPython