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

decision-circuits

Zero-dependency Python SDK for deterministic gates over calibrated model answers, with LangChain, OpenAI and Claude agent hooks.

LibrarybetaMIT
View on GitHub ↗

Overview

decision-circuits lets you ask a model small typed questions about a piece of state, get probabilities back, and then decide in code with thresholds, AND/OR/NOT, argmax with a confidence floor, majority votes, verification and ordinal buckets. That code is the decision circuit: the model never sees it, so it is versionable, testable offline and auditable, and each result carries its probability, outcome and trace. When a probability sits too close to a threshold, the gate abstains or escalates instead of guessing.

Any backend with one method can answer; a SystemOne backend targets TypeSafe's Jev. Integrations include LangChain tool-guard middleware and a model router, OpenAI Agents SDK input, output and tool guardrails, and a Claude Agent SDK pre-tool-use hook returning allow, deny or ask. It installs with no dependencies on Python 3.10+.

What the decision model decides

TypeQuestionAnswers
Yes / noDoes this text contain personal information about a private individual?yes · no
Yes / noIs the writer angry?yes · no
ChoiceWhich team should handle this?billing · technical · other

How it works

  1. Define questions and gates in a Circuit
  2. Run it against a backend such as Jev
  3. Gates compute decided, abstain or escalate outcomes with traces
  4. Use gates to redact, route, or send to a human

Strengths

  • Decision logic is versioned and testable offline
  • Explicit abstain/escalate near thresholds
  • No dependencies
  • Agent framework integrations

Caveats

  • Small project (early stage)
  • Calibration depends on the backend

Getting started

pip install decision-circuits, then build a Circuit and run it with decision_circuits.backends.SystemOne.

At a glance

Type
Library
Status
beta
Language
Python
Runtime
library
Has code
yes
Has tests
yes
License
MIT
Stars
1

Use cases and integrations

How curated lists describe it

  • Decision circuits: typed questions to a System One model, calibrated probabilities back, gates in code. Zero-dependency Python SDK with LangChain, OpenAI Agents, and Claude Agent SDK integrations.

    wh000wh000/awesome-jev-live

Frequently asked questions

What is a decision circuit?

Code that combines model probabilities with thresholds and logic to produce auditable decisions.

Do I need Jev?

No, any backend with the required method works; SystemOne is the Jev backend.

Can it guard agent tool calls?

Yes, via LangChain, OpenAI Agents SDK and Claude Agent SDK integrations.

Related projects

  • claude-jev-mod

    Claude Code mod that adds $.jev to the engine interface so other plugins and hooks can ask Jev typed questions through six providers.

    ServicebetaTypeScript
  • Ax

    Ax, a DSPy-style multi-language LLM framework, listed with a native TypeSafe provider for Boolean, Choice, Score and raw Jev questions.

    PluginproductionTypeScript
  • Composio

    Composio agent SDK with an optional TypeSafe provider that uses Jev to shortlist and pick tools and closed-set arguments.

    ProjectproductionTypeScript
  • Effect

    Effect TypeScript library whose @effect/ai-typesafe package maps classify, probability and rating operations to Jev questions.

    ProjectproductionTypeScript
  • instructor-php

    PHP monorepo whose Polyglot module adds a TypeSafe driver for typed Noul, Choice and Score decisions alongside LLM inference.

    ProjectproductionPHP
  • LangChain

    LangChain's Python framework, which includes an optional TypeSafeClassifier runnable plus model-routing and risky-tool middleware using Jev.

    PluginproductionPython