Skip to main content
The Chats section is the account’s activity in one list: conversations from every assistant, wherever they came from — the console, an embedded widget, the API, an inbound email or a trigger. Chats console

​
Finding a conversation

Filters narrow the list along the axes that matter in practice: There is also full-text search inside a conversation’s messages, for when you remember what was said but not where.

​
Reading one

Conversation detail A conversation shows more than the transcript:
  • Tool calls, expandable to their arguments and results — the fastest way to see why an answer was wrong.
  • Recalled memories, when the assistant used memory: what it brought in and from where.
  • Cost and tokens for the conversation.
  • Feedback left on individual messages, thumbs and comments.
When an answer is wrong, the tool calls usually explain it before the prompt does. A tool that returned an error, or was never called at all, is a different fix from a prompt that asked for the wrong thing.

​
Conversations that were blocked

A conversation stopped by a usage limit is marked as such, with the rule that blocked it, the current value, the ceiling and when the window resets. The mark clears itself as soon as a later message goes through.

​
Agent runs

Agent threads have their own equivalent view, with the run’s state, its tasks, approvals and pauses. See Logs.

​
Through the API

Everything here is available programmatically — list conversations, read one, search inside it and submit feedback — which is how you build your own review workflow or export the lot.