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

Run the decision model where the data already is. Same request, same answers, your infrastructure.

Every decision the model makes is made on your tickets, incidents, commands and documents. For the teams the benchmark was written for, “where does it run” is the first question, not the last. The open encoders run on four CPU cores at 150–630 ms per case; the same image runs on a GPU at tens of milliseconds.

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

A hosted judge is a data processor

Guardrails, triage and routing see everything.

Latency includes the internet

A 40 ms decision behind a 300 ms round trip is a 340 ms decision.

Zero-shot cannot be tuned

Hosted-only means nobody trains on your traffic — not even you.

How it works with Decide
  1. 1

    Pull the image

    The same container we host. Start with the general checkpoint.

  2. 2

    Point your client at it

    A base URL. The request and the answers are the same.

  3. 3

    Tune and calibrate locally

    Your labelled decisions never leave; the checkpoint and the temperatures are yours.

docker run -p 8000:8000 ghcr.io/metatext/decide-1:latest
client = SystemOne(base_url="http://decide.internal:8000")   # same request shape
  • 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

Benchmark timings, clean single-job, four CPU cores, 3-question case: open encoder 378 ms; GLiNER2-large 627 ms; hosted Jev ~450 ms including the round trip.

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

GPU required?
No. The models are 150M–450M parameters. A GPU makes them ~10× faster.
What about the registry and MCP server?
Both ship in the same image on the Enterprise plan; the hosted registry is the default elsewhere.

Related