Reference

New here? Start with the quickstart.

decide asks typed questions about your data and saves every answer. Point it at files, a folder, a JSON export, or piped text. It shows each answer with its probability, and keeps the results so you can look at them again. The CLI is experimental. Commands and file formats may change before v1.

Three ideas

• A template is a set of questions, such as “Is this file risky?” or “Which queue does this ticket belong to?”. decide templates lists them.
• decide run TEMPLATE DATA asks the template’s questions about each item in your data.
• Every run is saved. decide runs lists past runs, and decide runs view shows the latest results again without calling the model.

Reading the output

Each question has one of three types, and its answer has a probability:

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 Clear responsibilities…

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 being flagged is yellow, and the others are green. When a run has more than one item, the summary lists the flagged ones:

marker/app.py
! risk             yes           82%
  maintainability  ━━━━━━━─────  2.4 of 4  Understandable with some friction

✓ 5 answered  ! 2 flagged  500ms
Flagged: marker/Makefile, marker/app.py

Other templates look for matches instead, such as the items relevant to your question. A match is bold green and marked with ●, the other answers are dim, and the summary lists the matches when there is more than one item:

notes.txt:12  Annual plans are now 20% cheaper than monthly ones
● relevant  yes     91%

✓ 30 answered  ● 4 matched  1.1s
Matched: notes.txt:12, notes.txt:15, notes.txt:22, notes.txt:28

An answer to a question with no flag or match is shown in cyan: it is a value, not a verdict. 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.

In a terminal, decide shortens a long description or preview with … so each line fits the window. Piped output keeps every line whole. Add --details to see the full probability of every option and the model’s confidence, or choose another output format.

Pages

Page	Covers
Commands and flags	every command and flag, saved runs, exit codes, and environment variables
Items	what one item is in each kind of data, and choosing what to read
Diffs	judging what changed, by hunk, function, file, or line
Templates	the built-in templates, and writing your own
Providers	Jev, Clef, GPT-6 Luna, and your own service
Output formats	JSON, CSV, Markdown, and GitHub annotations
The answer cache	when decide asks again, and when it doesn’t
Claude Code	the plugin, the audit, and the hunt

For the Go library, see the package documentation.