Skip to main content
Some endpoints in this section are in beta and may change. Each endpoint’s page says whether it’s in beta.
These endpoints drive an app’s AI chat. Send a message and the AI builds from it, read the conversation, answer the questions the AI pauses on, and roll back changes you don’t want. Sending a message is how you build an app after you create it.

Turns

Each message starts a turn, which covers everything the AI does in response. A turn can take several minutes on a large change, and the request stays open until it settles. Each turn consumes credits. See Apps API concepts for how a turn ends. A 200 response doesn’t mean the turn succeeded, so read the result before you treat it as done. To retry a request without paying twice, see Retrying chat requests.

When the AI waits for you

Some steps pause for approval. A tool call with status set to waiting_for_user_input means the AI can’t continue until you answer it. Find it with Read conversation messages, then answer it with Submit tool-call input, or answer several at once with Submit tool-call input (batch).

Change course

  • Stop generation stops the turn that’s running. Changes already saved as a checkpoint are kept.
  • Undo message rolls the app back to the checkpoint a message produced.
  • Edit and resend replaces an earlier message and rebuilds from there, so the changes after it are undone.

Endpoints

To retry a request safely, see Retry chat requests.