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

instinct-ai

Open-source dual-brain runtime that answers typed decisions with a fast local engine or the Jev API before falling back to LLMs.

ServicebetaApache-2.0
View on GitHub ↗

Overview

Instinct AI (also importable as reflex) presents itself as a universal System One runtime and gateway. Its argument is that using large autoregressive models to make boolean and routing decisions wastes time and money, so it offers Noul, Choice and Score primitives evaluated in a single pass, auto-detecting a local engine the README describes as sub-15 ms or the cloud TypeSafe Jev API, with heavier reasoning models as a second brain.

It is aimed at Python and JavaScript developers who want typed decisions behind one API, and at agent users through an MCP server for Claude Desktop, Cursor and other MCP clients. It includes inline shortcuts such as ins.noul and ins.choice, a browser playground that compares System One and System Two side by side with telemetry, and gateway statistics like cache hit rate and dollars saved. Packages are published on PyPI and npm, with CI and an Apache-2.0 licence per the badges.

What the decision model decides

TypeQuestionAnswers
Yes / noIs the customer demanding a refund or chargeback?yes · no
ChoiceSelect operations queuebilling · technical_support · fraud_investigation · spam
ScoreCustomer distress score 1-10—

How it works

  1. Call ins.evaluate with state and typed questions
  2. Runtime routes to local engine or Jev API
  3. Typed results return with latency and cost
  4. Harder prompts can be escalated to a reasoning LLM

Strengths

  • One API over local and hosted decision backends
  • MCP server exposes decisions to Claude Desktop and Cursor
  • Playground visualises latency and cost versus an LLM

Caveats

  • Performance claims (sub-15 ms, $0) are the project's own and not independently verified
  • Local engine quality versus Jev is not documented in the excerpt

Getting started

Install from PyPI or npm (instinct-ai), then `instinct playground --port 8000` to try the dashboard.

At a glance

Type
Service
Status
beta
Language
Python, TypeScript
Runtime
library, gateway, MCP server
Has code
yes
Has tests
yes
License
Apache-2.0

Use cases and integrations

How curated lists describe it

Frequently asked questions

What does Instinct decide?

Any Noul, Choice or Score question you pass, such as refund requested or which queue to route to.

Does it need Jev?

It can use a local engine or the cloud Jev API, auto-detecting which is available.

Can agents use it?

Yes, through its MCP server for Claude Desktop, Cursor and other MCP clients.

Related projects

  • jev-x-kit

    MCP server, CLI and Claude Code skill with Choice, Score and Noul primitives and a confidence gatekeeper, hosted or offline.

    MCP serverbetaTypeScript
  • decidealot

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

    MCP serverbetaPython
  • 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-switch

    Unofficial Rust gateway routing /v1/systemone requests to TypeSafe, Vercel, OpenRouter or local Laya via an editable DAG.

    ProxybetaRust
  • noulxp

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

    ProjectbetaPython
  • one-system

    Go gateway that puts local and hosted System One models behind one TypeSafe API, with a local Laya router choosing Laya or Jev.

    ServicebetaGo