What is the Decima model?

A small local decision model that takes a situation, a question and options, and returns calibrated probabilities in milliseconds on a CPU. It is the judge, not the writer: put it next to an LLM or agent to decide safety, truth or routing.

Is Decima the same as TypeSafe Jev?

No. It is an independent Jev-style model that speaks TypeSafe's /v1/systemone format, so Jev client code works after a one-line change; TypeSafe publishes the official Jev models, and the README describes Decima as a Jev alternative.

How does Jev-compatible code talk to it?

Because it implements the same choice, noul and score protocol, the official typesafe-sdk runs unchanged against it: TypeSafeClient(api_key='local', base_url='http://127.0.0.1:11436') after starting decima serve.

What does it cost to run?

Nothing beyond your hardware: int8 ONNX runs about 20-50 ms per decision on one CPU core, with no GPU and no API key, and no data leaves the machine. The package itself is Apache-2.0 licensed.

What presets are available?

triage, guardrails, moderation, routing, pii, agent-gate and kg-judge, per decima presets. For example, agent-gate returns a deny/no gate for a destructive command, and triage returns category, urgency and sentiment.

Can I use any labels I want?

Yes. Options are your own: any labels, any number, supplied at call time, and the README says shuffling the options never changes the answer (0% flips, versus 10-27% for other open models in its tests).

How do I integrate it with agents?

Besides the CLI and Python Router, there are Claude Code hooks for secret and command gates with rules first and fail-closed behavior, an MCP server via decima mcp, LangChain/LangGraph integrations, Docker and a Colab quickstart.

Jev directory

CommunityJev Models & Reimplementations

Decima

Open, CPU-sized Jev-style decision model: give it a situation, a question and options, and Decima returns calibrated probabilities in roughly 20-50 ms per decision as int8 ONNX, with no GPU, no API key and nothing leaving the machine.

See highlights Open github.com

Stars: 1
Language: Python
License: APACHE-2.0
Reading time: 2 min
Published by: Community
Author: amyrmahdy
Last verified: Oct 2026

amyrmahdy/decima

User repository on GitHub

View on GitHub

Open, CPU-sized Jev-style (System One) decision model: situation + question + options → calibrated probabilities. 122M, ~20 ms on one CPU core, runs in your browser.

Forks: 1
Last pushed: Updated Oct 2026

Repo stats from the GitHub API, cached Oct 2026.

Tagscommunity catalog-import cobanov onnx local calibration

Works forJev for Classification Jev for Routing Jev for Decision gating

Used inJev for Software & DevTools

Languages: Python
Platforms: CLI MCP API Browser
Capabilities: Open sourceRuns locally

Highlights

122M-parameter model speaking TypeSafe's /v1/systemone format (choice, noul, score), so code written for Jev works after a one-line change.
The official typesafe-sdk works unchanged against decima serve: TypeSafeClient(api_key='local', base_url='http://127.0.0.1:11436').
CLI presets cover triage, guardrails, moderation, routing, pii, agent-gate and kg-judge; decima mcp runs it as an MCP server.
The README claims calibrated answers (0.8 is right about 80% of the time), zero answer flips when options shuffle and evaluation in 20 languages.
Integrations include Claude Code hooks with rules first and fail closed, LangChain/LangGraph, Docker and a Colab quickstart.

Quickstart

bash

pip install decima-ai
decima "I was charged twice for my subscription" --question "Which team?" --choices billing,technical,sales

Watch out

Copy reviewed against the truncated repo README on 2026-10-15; the excerpt stops inside the presets table. Calibration, order-proofness, latency and multilingual figures are the project's own README claims, and Decima is Jev-compatible rather than an official TypeSafe project.

Questions about this entry

Short answers drawn from the linked project or page.

From the Jev guides

Original write-ups on what this entry is for.

Exporting Decision Models: GGUF, ONNX, and Device Runtimes

Oct 10, 2026

Taking a decision checkpoint out of Python: what the GGUF and browser WebGPU exports preserve, what they change, and the parity evidence the model cards publish.

Jev Tools & CLIs#open-models#local#onnx

Read article

Beyond text and boxes: multimodal decisions you can run yourself

Oct 16, 2026

Jev-Omni's merged classifier scores text, image, audio, and video options; VideoJev and Laya Vision show what open multimodal decision models accept, what they return, and which numbers are the authors' own.

Jev Models & Reimplementations#multimodal#open-models#vision

Read article

Jev 1.14: Images, Box, and the Danger Zone

Oct 15, 2026

Jev Day shipped a routine model bump and a deliberate widening of the API: images on both models, a box question type on 1.14 only, and 262k context, all behind danger-zone headers.

Jev Models & Reimplementations#release#vision#danger-zone

Read article

From the Jev AI model community

Posts from builders shipping with Jev right now.

Follow @typesafeai

Decima

Highlights

Quickstart

Questions about this entry

What is the Decima model?

Is Decima the same as TypeSafe Jev?

How does Jev-compatible code talk to it?

What does it cost to run?

What presets are available?

Can I use any labels I want?

How do I integrate it with agents?

Related terms

More Jev resources like this

From the Jev guides

From the Jev AI model community