Vercel AI CLI
A terminal tool exposes typed Boolean, Choice and Score evaluations with Jev as the default evaluation model.
Added to Jevfast
How it uses Jev
The `ai evaluate` command reads text or JSON state from stdin and accepts named Boolean, Choice, or Score questions through flags or a JSON file. Jev is the default evaluation model; the CLI returns typed answers and evaluation metadata as JSON.
What Jev decides
Classify a refund requestExample answers · not a recorded Jev response · Source ↗
Question 1 · team
YOUR APP
INSTRUCTION
Which team?
STATE
Text/JSON supplied on stdin; the example evaluation asks whether a refund is requested, which team should handle it, and how positive the tone is.
JEV · CHOICE
- billing
- support
Question 2 · tone
YOUR APP
INSTRUCTION
How positive?
STATE
Text/JSON supplied on stdin; the example evaluation asks whether a refund is requested, which team should handle it, and how positive the tone is.
JEV · SCORE
0
App workflow
- Define questions
Provide one or more typed questions with CLI flags or a JSON question file.
- Pipe in state
Send text or structured JSON on stdin; the command parses it without flattening arrays.
- Evaluate
The CLI resolves the selected evaluation model, defaulting to Jev, and sends the state and questions through AI Gateway.
- Read JSON
The command prints the returned answers and evaluation metadata for downstream shell or application logic.
Why it is interesting
It makes typed judgments usable from shell pipelines and scripts while leaving the question definitions and downstream interpretation under the caller's control.