Questions and answers

A decision model answers a question with yes or no, a choice, or a score, instead of generating text. decide asks it about each item in your data, and each answer comes back with its probability, so a script or a program can act on it directly. There is no prose to parse.

Yes or nonoul

Could this command destroy work that is hard to get back?

!destructiveyes93%

git push --force origin main

Flagged at yes >= 80%

One of severalchoice

Which team should take this support ticket?

queuebilling91%

billing91%engineering7%other2%

Route it with --format json

A point on a scalescore

How maintainable is this file?

maintainability2.0 of 4

━━━━━━──────

Understandable with some friction

Flagged at <= 1.5

Templates, items, and runs

• A template is a named set of questions, such as “Is this file risky?” or “Which queue does this ticket belong to?”. decide templates lists the built-in ones, and decide templates new starts your own.
• An item is one thing a template’s questions are asked about: a file, a changed function, a JSON record, a line of text, or an image. decide run TEMPLATE DATA asks about each item in your data.
• A run is saved with every answer. decide runs lists past runs, and decide runs view shows the results again without asking the model.

Three types of question

Type	Asks	Shown as
noul	a yes-or-no question	risk no 88%, or unsure 52% yes between 40% and 60%
choice	which one of several options fits	queue billing 91%
score	where something falls on a scale	maintainability ━━━━━━━━──── 2.7 of 4

A noul answer is the probability of yes. A choice answer is the most likely option, with the probability of every option. A score answer is a point on the scale, with the probability of every level.

Flags and matches

Some templates flag the answers that need attention, such as a file that is probably risky. A flagged answer is red and marked with !, an answer close to a flag is yellow, and the others are green. The summary lists the flagged items, and --fail-on flagged exits with code 2 when there are any.

Other templates look for matches instead, such as the items relevant to a question you choose. A match is bold green and marked with ●, and --fail-on matched exits with code 2 when there are any.

decide templates show TEMPLATE says when each question is flagged or matched. A yes-or-no answer between 40% and 60% is yellow in every template: the model is unsure.

Providers

decide runs on three decision models, with the same commands, templates, and Go code:

Model	By	Runs on
jev-latest, the default	TypeSafe	the TypeSafe API
clef, and clef-flash for lower latency	Cloudflare	Workers AI
gpt-6-luna	OpenAI	the Decisions API, in beta

Any other service that speaks the Jev API works too. The quickstart shows how to set a key for each.