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

Foq

Local open alternative to Jev using a 1.58-bit ternary 8B model to answer booleans, choices, scores and Pydantic extraction in ~25 ms.

ProjectbetaMIT
View on GitHub ↗Live demo ↗

Overview

Foq presents itself as a free, open-source, fully local alternative to Jev. It is a decision engine for AI agents that runs one feed-forward pass over the context and returns calibrated probabilities on typed questions (booleans, choices, scores and Pydantic objects) without generating prose, in about 25 ms on your own machine according to the README.

The reference model is Ternary Bonsai 8B by PrismML, a Qwen3-8B architecture with 1.58-bit ternary weights running in C++ at about 4.8 GB of memory, with Foq adding a decision calibration layer. The README compares Foq favourably with Jev and Laya, including a claim of 100% on its operational suite, but these are the author's own comparisons. Code is MIT and the model is Apache 2.0; Foq pins a verified copy by hash and does not redistribute weights.

What the decision model decides

TypeQuestionAnswers
Yes / noIs this action destructive?true · false
ChoiceWhere should this go?auto · human

How it works

  1. pip install foq and create a FoqEngine
  2. Call system_one with state and typed questions
  3. Read answers with confidence
  4. Use extract() for Pydantic schema outputs

Strengths

  • Fully local and free
  • Supports structured extraction
  • Small memory footprint for an 8B model

Caveats

  • Benchmark claims are the author's own
  • Marketing-style README

Getting started

pip install foq, then FoqEngine().system_one(state=..., questions={...}).

At a glance

Type
Project
Status
beta
Language
Python
Runtime
local engine (C++ backend)
Has code
yes
License
MIT
Stars
11

Use cases and integrations

How curated lists describe it

  • ⚡ Foq — the FREE, local, open-source alternative to Jev. Typed System 1 decisions in ~25 ms — no waitlist, no cloud, no per-token cost. foq.fr

    wh000wh000/awesome-jev-live

Frequently asked questions

Is Foq related to TypeSafe?

No, it is an independent local alternative.

Which model does it use?

Ternary Bonsai 8B by PrismML with a Foq calibration layer.

Does it need a GPU?

The README describes about 4.8 GB of memory; it runs on your own machine.

Related projects

  • Decima

    122M multilingual local decision model answering Choice, Noul and Score in milliseconds on CPU behind a /v1/systemone server.

    BenchmarkbetaPython
  • brier

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

    ProjectproductionPython
  • blink★ 22

    Open one-pass typed-decision model with a dependency-free C99 runtime that also builds to a 66 KB WebAssembly module.

    ServicebetaC
  • browser-laya

    Runs the open Laya decision model entirely in a browser tab with WebGPU and ONNX Runtime Web: no server, no API key.

    ServicebetaTypeScript
  • ifReflex.cpp

    llama.cpp-based C++ server that turns any LLM into a System One decision model, reading answers from next-token logits.

    ProjectbetaC++
  • jevalt

    Three open 4B decision models for English, Turkish and German that serve the Jev /v1/systemone API on a CPU in about 3 GB of RAM.

    ServicebetaPython