Skip to main content
ADE 1.2.123 keeps each chat’s model in the composer, removes the Fast button from the prompt box until you turn Fast mode on, and sends fewer false error reports.

​
Improved

  • The prompt box no longer shows a Fast button when Fast mode is off for the account. A FAST chip in the model picker opens a short dialog that takes you to Settings › Chat › “Allow Fast mode” (#1896).
  • The model hint no longer suggests a weaker model from your past picks alone. Only a clearly small request can get a cheaper-model suggestion (#1895).
  • Expected errors (no git repository, a refused path, a network drop) stay in local logs and no longer create error reports (#1894).

​
Fixed

  • A chat that already ran never shows “Select model”. The composer reads the model from the chat summary, then from the transcript. A chat’s model no longer changes by itself when you send (#1894).
  • A window recovers when the brain no longer knows its project id, without a restart (#1894).
  • Long chats on Windows no longer run the window out of memory (#1894).
  • The Issues pane “Browse Jira” crash, a clear error for chat.updateSession with no arguments, and a new ade chat update command (#1894).
  • The web app reloads once when a deploy removes a script it needs (#1894).

​
iOS

  • A sign-in that the machine refuses with a clear next step no longer creates an error report (#1894). This is in a new TestFlight build.