pyjev

Confidence-aware Jev decisions for Python, shell, CI, and automation.

pyjev is a thin operational layer over TypeSafe's official typesafe-sdk. It preserves Jev probabilities, confidence, and response metadata while adding reusable named decisions, shell-safe output, credential ergonomics, local validation, request inspection, and confidence gates.

Install

pip install pyjev
# or: uv tool install pyjev

For a development checkout:

pip install -e '.[dev,docs]'

A first decision

pyjev choice "Where should this ticket go?" \
  --state "Stripe checkout fails" \
  --option billing="Payments and refunds" \
  --option engineering="Technical failures" \
  --option sales="Purchasing questions" \
  --json

from pyjev import Jev

with Jev() as jev:
    result = jev.choice(
        "Where should this ticket go?",
        state="Stripe checkout fails",
        choices={"billing": "Payments", "engineering": "Technical", "sales": "Purchasing"},
    )
print(result.value, result.confidence, result.probabilities)

What pyjev adds

• direct Noul, Choice, and Score operations with uncertainty preserved;
• reusable .pyjev.toml decisions and mixed-question bundles;
• offline validation and request compilation;
• confidence-gated shell automation and stable exit codes;
• keyring, environment, and explicit plaintext-file credential support;
• request IDs, usage, raw metadata, and structured JSON for debugging.

The official typesafe-sdk remains responsible for transport, retries, API models, authentication at the HTTP layer, and API error semantics. pyjev does not invent probabilities or confidence calculations.

Learn more

• Full documentation
• Runnable examples
• TypeSafe Jev documentation
• Issues

License

Apache-2.0