Steadier agents and clearer errors
Conversations recover when an agent loses its session, answers wait for an offline computer, and errors say what to do.
A round of fixes on the server makes long-running work more dependable and failures easier to act on.
- When an agent loses its session, the next message opens a new one instead of failing.
- An answer to an agent's question waits when its computer is offline, and is delivered once it is back.
- Plain messages when a conversation is too long for the model, when an endpoint does not serve the chosen model, when Codex needs to sign in again, or when the plan has no room for another cloud computer.
- A Trigger that keeps failing for a lasting reason is paused instead of retrying forever, and scheduled Triggers are spread across their minute.
- Members who are not admins see only their own usage.