fuzzy-jev for Python

Python bindings for fuzzy-jev: ask TypeSafe's Jev typed questions about a piece of text through OpenRouter's decisions endpoint, and get probabilities back — a Choice between named options, a Score on an ordered scale, or a Noul, the probability that something is true. Rules then combine those answers by fuzzy logic into decisions and amounts, and draw themselves as SVG.

The package is fuzzy-jev; it is imported as jev. It is the Rust library underneath, so a question, a reply and a rules file mean exactly what they mean to the jev command.

Install

pip install fuzzy-jev

Wheels are built for Linux and macOS (x86_64 and arm64), for CPython 3.10 and later. Anywhere else pip builds from source, which needs a Rust toolchain. From a checkout of this repository:

pip install ./python          # or: uv pip install ./python

For development, maturin develop builds it into the active virtual environment.

Ask

import jev

client = jev.Client()  # the key from OPENROUTER_API_KEY; or jev.Client("sk-or-...")
reply = client.decide(
    "Help! My payouts have been failing for 3 days.",
    {
        "is_urgent": jev.Question.noul(
            "Does this message convey urgency?", yes="Explicitly time-sensitive", no="No urgency expressed"
        ),
        "department": jev.Question.choice(
            "Which team should handle this?",
            {"billing": "Payments, invoicing, refunds", "technical": "Bugs, outages, integrations", "sales": "Pricing"},
        ),
        "frustration": jev.Question.score("How frustrated is the customer?", ["Calm", "Frustrated", "Very angry"]),
    },
)

reply.noul("is_urgent")                  # 0.95
department = reply.choice("department")  # ChoiceAnswer: .choice, .confidence, .probabilities
reply.score("frustration").score         # 1.04, between levels 1 and 2
print(reply.render("table"))             # as `jev --table` prints it
reply.usage.cost                         # in US dollars

The state can be a string, a dict or a list; instructions and criteria can be any JSON value too. Questions are a dict of id to Question, or (id, question) pairs; a question can also be a dict in the endpoint's shape ({"type": "noul", "instructions": ...}), and jev.load_questions(text) reads a jev -q questions file.

decide releases the GIL while it waits, so threads ask in parallel; await client.decide_async(...) is the same for asyncio. client.request(state, questions) is the request body decide would send, without sending it.

Errors are all jev.JevError: StatusError (with .status), HttpError, DecodeError, InvalidQuestionError (raised before any call), MissingAnswerError, WrongTypeError, BadAnswerError and RulesError.

Rules

questions = jev.load_questions(open("examples/rules/triage.json").read())
rules = jev.Rules(open("examples/rules/triage.toml").read(), questions)  # checked now, before a call

reply = client.decide(ticket, questions)
outcome = rules.evaluate(reply)
outcome.yes                              # ["page on-call", ...]: the items at or over the threshold
outcome["reply_within"].value            # an output's crisp value, in hours here
print(outcome.text())

open("triage.svg", "w").write(rules.graph_svg(reply))

See docs/rules.md for the whole of the rules file.

More

The whole guide: questions, the reply, rules, drawings, errors, threads and asyncio, and releasing.

Tests

maturin develop && pytest