decision-toolkit

Local-first toolkit for open-weights decision models on CPU: typed choice / yes-no / score questions with calibrated probabilities, one ONNX engine and one adapter per model contract.

uv pip install decision-toolkit
uv pip install "decision-toolkit[onnx]"     # model runtime
uv pip install "decision-toolkit[server]"   # REST server

Development builds are reproducible with the committed uv.lock.

PyPI: https://pypi.org/project/decision-toolkit/

Start here

• Examples — library, REST and CLI.
• Core API — the superset used by every surface.
• REST — TypeSafe System One + OpenAI Decisions beta.
• CLI — operational commands and batch workflows.
• Variants and quantization — every catalogued build and how to download/select it.
• Operations — rate limits, metrics, tracing and runbook.
• Benchmarks — reproducible latency/parity/calibration.
• Models — the curated catalog.
• Release — build, TestPyPI/PyPI and Trusted Publishing.

One example

from decision_toolkit import DecisionModel, choice, noul

with DecisionModel("gliner2.5-multi") as model:
    decision = model.decide(
        "Meeting with Acme: 10% discount, sign next week.",
        {
            "cat": choice("What kind of record is this?", {
                "meeting": "a meeting or call",
                "tasks": "things to do",
                "other": "anything else",
            }),
            "urgent": noul("Does it need a reply today?"),
        },
    )

decision["cat"].choice
decision["cat"].confidence
decision["urgent"].noul
decision.usage.to_dict()

Trusted publishing

Every release from 0.1.2 onward is published through GitHub OIDC to TestPyPI and PyPI. See RELEASE.md.