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

A Jev alternative with the same request shape — open weights, tuned to your decisions, run where your data lives.

Jev created the category: one call, typed questions, calibrated answers, nothing generated. The objections to Jev are what people search for next — hosted-only, no registry, no extraction, no way to tune it on your own traffic. That list is this product.

Get early access See the benchmark

In development — the benchmark numbers on this page are measured; the Decide-1 model is in training.

The problem

Your data has to leave

Jev is hosted-only. Every ticket, incident and trace you decide on goes to a third party. For many teams that is the end of the evaluation.

Zero-shot is the ceiling

There is no way to teach Jev your domain. On a public 2,000-decision suite it scores 0.726 and stays there; a fine-tuned open encoder reaches 0.714 and 0.762 after calibration.

Every question, every request

Jev has no registry: the same three questions are redefined inline in every service that asks them, and drift.

How it works with Decide
  1. 1

    Change the base URL

    The request is the one you already send: `POST /v1/systemone` with `{state, questions}`. `choice`, `score` and `noul` come back in the same shape. A Jev integration moves in one line.

  2. 2

    Send a few hundred of your decisions

    Labelled or teacher-labelled. You get back a checkpoint tuned to your workflow and one fitted temperature per question type — the two things the benchmark shows close the gap.

  3. 3

    Run it where you want

    Hosted by us, or the same container in your VPC. Apache-2.0 weights underneath; no lock-in either way.

# Same request shape as Jev — only the base URL changes
client = SystemOne(base_url="https://api.metatext.io", api_key=...)
r = client.system_one(
    state=ticket,
    questions={
        "department": {"type": "choice", "instructions": "Which team?",
                       "criteria": {"billing": "...", "technical": "...", "sales": "..."}},
        "refund_requested": {"type": "noul",
                             "instructions": "The customer explicitly asks for a refund."},
    },
    model="decide-1-latest",
)
r.answers["department"].probabilities   # sums to 1, calibrated
  • Choice — pick one option, a probability on each
  • Score — a position on your ordered rubric
  • Noul — a calibrated yes/no probability
  • Extract — spans and records with offsets
Something you can check

On the public typed-decisions suite (2,000 decisions, one harness, raw JSON published): Jev 0.726 zero-shot; an open encoder fine-tuned on the workflow 0.714, and 0.762 held-out after one temperature per question type.

Numbers about Jev are its authors' or the cited source's; numbers from the benchmark are ours, on one harness, with the raw JSON published. Full table.

Get early access — and tell us which decision you'd hand over first

We're building Metatext Decide with the first teams that use it. Leave your email, answer as much as you like, and say yes to a short interview if you want a say in what ships.

  • Early access to Decide as it opens up
  • A 20-minute interview if you want one — research, not a pitch
  • The benchmark harness and raw results, today
  • No spam — early-access updates only
Where do decisions happen in your stack today?
What kind of decisions?
Have you tried Jev?
What stops you using a hosted decision API?
Would you pay?
Open to a 20-minute interview?

Research, not a pitch — we want to hear how you decide today.

Early-access updates only. No spam, unsubscribe any time.

Questions

Is this affiliated with TypeSafe or Jev?
No. Same request shape, different models, no shared data or code. We benchmarked the real Jev on our harness and publish the numbers, including the ones where Jev wins.
Does an open model match Jev zero-shot?
No — not on any suite we have run. It matches after fine-tuning on the target workflow, which is the product.
Can I keep using Jev for some questions?
Yes. The request shape is identical, so routing by base URL is trivial.

Related