# pi-system-one

> A Pi and OMP extension that exposes one provider-neutral `system_one` tool for bounded AI decisions.

Canonical: https://iamaamir.github.io/system-one/pi-system-one/
Source: https://github.com/iamaamir/system-one/tree/main/pi-system-one
npm: https://www.npmjs.com/package/pi-system-one

## What it does

`pi-system-one` adds one `system_one` tool to Pi and OMP.

The tool supports:
- `choice`: choose among unordered explicit alternatives.
- `noul`: return a calibrated probability that a yes/no proposition is true.
- `score`: place a judgment on an ordered rubric.

One call can batch heterogeneous bounded judgments over shared state.

## Providers

TypeSafe Jev is the tested hosted example.

Configuration:
- `SYSTEM_ONE_BASE_URL=https://api.typesafe.ai`
- `SYSTEM_ONE_API_KEY=<TYPESAFE_API_KEY>`
- `SYSTEM_ONE_MODEL=jev-latest`

Any compatible System One model, endpoint, or provider can sit behind the same `system_one` interface.

## Install

Pi:
`pi install npm:pi-system-one`

OMP:
`omp install npm:pi-system-one`

## When to use it

Use the normal agent to find facts or create open-ended content.

Use `system_one` after the evidence is available and the remaining task is bounded: choose, judge yes/no, or score.

These are examples, not routing rules. In normal use, the agent can choose
`system_one` when the task has narrowed to a bounded choice, yes/no judgment,
or score. `/judge` invokes that judgment workflow explicitly.

Coding-agent examples:
- review a completed agent run and decide whether to auto-close, request human review, file an issue, or page on-call
- triage a failed test as a regression, flaky test, environment/tooling failure, or specification mismatch
- accept, revise, or escalate a patch after deterministic checks
- score a concrete code-review finding by severity
- judge whether an agent trace attempted to cross its granted permission boundary
- choose among candidate fixes after tests and benchmark evidence are available

## Explicit shortcut

`/judge` explicitly invokes the System One judgment workflow when you do not want tool selection to depend on agent inference.

## Related

System One core:
https://iamaamir.github.io/system-one/

Pi-Bifrost:
https://iamaamir.github.io/pi-bifrost/