Skip to content
Metatext Decide — typed decisions for your agents, with a probability on every answerGet early access
Metatext
Sign Up
CI & code review

Is the migration reversible? Is an error swallowed? Does the PR description match the diff? Ask typed questions in CI.

The jev-code project gives coding agents five tools — classify, check, score, rank, ask — and teams point them at CI: grade a PR, triage a wall of failures, enforce conventions that are semantic, not syntactic. A typed check costs a fraction of a cent per pull request hosted, and nothing self-hosted.

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

Conventions live in review comments

“We don’t swallow errors” is a rule nobody can lint.

Forty failures, three real

Flaky, infra, real — triage is a choice question, made by hand.

Descriptions drift from diffs

Nothing checks that the PR says what the code does.

How it works with Decide
  1. 1

    Encode the conventions

    One noul each; criteria written by the reviewer who usually catches it.

  2. 2

    Score severity

    A 0–3 score decides comment vs block.

  3. 3

    Triage failures

    Choice over real / flaky / infra per failure, batched in one call.

checks = {
  "reversible":  {"type": "noul", "instructions": "The migration can be rolled back safely."},
  "swallowed":   {"type": "noul", "instructions": "An error is caught and ignored."},
  "desc_match":  {"type": "noul", "instructions": "The PR description matches what the diff does."},
  "severity":    {"type": "score", "instructions": "Worst issue severity", "criteria": ["none","nit","should fix","blocking"]},
}
r = client.system_one(state={"diff": diff, "description": pr.body}, questions=checks)
  • Noul — a calibrated yes/no probability
  • Score — a position on your ordered rubric
  • Choice — pick one option, a probability on each
Something you can check

Pattern from jev-code (MCP server + skill for Claude Code, Codex, Pi, OpenCode) and the maximem.ai decision-seams map: “fractions of a cent per pull request” on hosted Jev.

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

Will it replace code review?
No. It turns the checks reviewers repeat into numbers, so reviewers spend time on the rest.

Related