Skip to main content
The chat splits everything it can do into two kinds. Reading — listing tests, checking a run, searching sources — happens immediately, no permission needed. Changing state — creating, editing, deleting, running — is proposed first: the chat posts a card describing exactly what it wants to do, and nothing happens until you resolve it.
A run proposal — the card names the environment and target URL, and waits for Run or Reject

​
What a card shows

  • The action and its full payload — the exact values that will be used, not a summary. A run card names the environment it will hit; a delete card spells out what is destroyed (“project X, its 43 test cases, and their run history”).
  • Pre-filled fields you can correct before approving, where the action supports it.

​
Three ways to resolve a card

One typed sentence can settle several open things at once — you’re never forced through cards one field at a time.

​
Question cards

When the chat needs information it can’t infer — a URL, a choice between options, a credential — it asks with a question card: single/multi select, free text, or password. Multiple questions render as a stepper (“1 of 3”), and selects include an “Other” option where a custom answer makes sense.
Secrets are always collected through password fields — masked, never echoed back into the conversation, and never readable back out of a saved configuration (you’ll see password set instead of the value). Leading/trailing spaces are kept, since they can be part of a secret.

​
Where to Go Next

AI Chat Overview

What the chat can do and how project focus works