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

Which skill fits this request? Rank up to 255 options in one pass and load only when one clearly applies.

TypeSafe published the cleanest study in the category: on 488 requests, putting a ranking decision in front of skill loading cut wrong-skill loads from 16.8% to 7.3% and loads-when-nothing-fit from 9.8% to 4.0%. Every loaded skill is context; every wrong one is wasted context.

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

The agent loads by name

Skill descriptions are matched by whatever the model feels like; nothing scores the alternatives.

Nothing fits, something loads

One in ten loads happened when no skill applied at all.

Rosters grow

Fifty skills is a long prompt. A choice over 255 options is one forward pass.

How it works with Decide
  1. 1

    Stage one: the roster

    Choice over every skill name + one-line description; noul “a skill applies”.

  2. 2

    Stage two: the top three

    Re-rank with full descriptions; load when the winner clears 0.6 and the noul clears 0.5.

  3. 3

    Log and tune

    Whether the loaded skill was used is a label. Tune on it.

stage1 = client.system_one(state=request, questions={
  "skill": {"type": "choice", "instructions": "Which skill fits?", "criteria": ROSTER},   # ≤ 255
  "any":   {"type": "noul",   "instructions": "At least one skill applies."},
})
if stage1["any"].noul < 0.5: load_nothing()
else: load(rerank(top3(stage1["skill"].probabilities)))
  • Choice — pick one option, a probability on each
  • Noul — a calibrated yes/no probability
Something you can check

TypeSafe’s study (488 requests, claude-haiku-4-5 agent, jev-1.12 ranking): wrong-skill loads 16.8% → 7.3%; loads when nothing fit 9.8% → 4.0%; theoretical floors 2.5% / 1.2%.

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

Does this work with MCP tools as well as skills?
Same mechanism: the roster is a list of names and descriptions, whatever they point at.

Related