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

opendecider

Open calibrated decision models from 400M to 80B with Jev-compatible server, MCP server and clients, benchmarked against Jev.

ModelbetaApache-2.0
View on GitHub ↗Live demo ↗

Overview

opendecider provides open, calibrated System 1 decision models: ask choice, score and noul questions about text, email, tickets or JSON and get a calibrated probability per option, with the README citing 17 ms on an NVIDIA GPU and 18 ms on a Mac for the small model. Models range from about 400M to 80B (per the list summary), are distilled from open teachers, and are benchmarked head-to-head against TypeSafe Jev, Microsoft-Decision-1, Laya, CLM-8B and frontier LLMs with the same scorer.

Packaging is broad: pip extras for nano and small models, MLX builds for Apple Silicon, a Jev-compatible /v1/systemone HTTP server, an MCP server for Claude Code, Claude Desktop and Cursor, a PyTorch-free client with routers and a guard, a TypeScript client with Vercel AI SDK and Mastra tools, and GGUF builds for LM Studio, Ollama and vLLM. It runs on CPU, CUDA and Apple Silicon, is Apache-2.0, and has a Colab notebook and demo space.

What the decision model decides

TypeQuestionAnswers
ChoiceWhich option fits this state?—
ScoreScore the state on ordered levels—
Yes / noIs this statement true?yes · no

How it works

  1. pip install opendecider (or a variant).
  2. Run in-process, via the /v1/systemone server or the MCP server.
  3. Ask typed questions and get calibrated probabilities.
  4. Use the router or guard helpers to act on them.

Strengths

  • Wide range of sizes and runtimes.
  • Jev-compatible server plus MCP and TypeScript clients.
  • Published head-to-head comparisons.

Caveats

  • Comparisons are author-run.
  • Larger models need significant hardware.

Getting started

pip install opendecider (add [serve] for the /v1/systemone server or [mcp] for the MCP server).

At a glance

Type
Model
Status
beta
Language
Python
Runtime
library, HTTP server, MCP server
Has code
yes
Has tests
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

  • Calibrated decision models from 400M to 80B on CPU or NVIDIA, with weights on Hugging Face and a Colab notebook.

    MrJev/awesome-jev

Frequently asked questions

Is it compatible with Jev clients?

Yes, the serve extra exposes a Jev-compatible /v1/systemone endpoint.

Can I run it on a Mac?

Yes, including MLX 4-bit and 8-bit builds of the small model.

Which agents can use it via MCP?

The README names Claude Code, Claude Desktop and Cursor.

Related projects

  • jev-mcp

    MCP server exposing Jev-style decision models served locally by Ollama's /v1/systemone as classify, yes/no and score tools.

    MCP serverbetaTypeScript
  • jev-rs

    Rust engine that answers noul, choice and score from any LLM in one prefill via logprobs, wire-compatible with /v1/systemone and MCP.

    ModelbetaRust
  • decidealot

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

    MCP serverbetaPython
  • noulxp

    NoulXP: open standard and reference engine for packaging calibrated System One decision models so one runtime can run them all.

    ProjectbetaPython
  • arbiter

    Serving layer that runs typed-decision models like Laya on your GPU or Mac behind a Jev-compatible /v1/systemone API.

    ServicebetaPython
  • brier

    Mangaba AI's local, calibrated typed-decision model for Portuguese text, trained to optimize the Brier score; proprietary since v5.

    ProjectproductionPython