Queue, fork and restart
See why a chat waits, manage its queued messages, and fork or restart it.
Why a chat waits
A chat that cannot run yet says why, in one line above its composer:
| What it says | What it means |
|---|---|
| All 3 of your agent environments are in use. 2nd of 4 in line. | Your plan runs a set number of chats at once. It starts as soon as one finishes, in the order shown. |
| During the Founding period, each seat runs up to 10 agents at once. This one starts as soon as one finishes. Next in line. | Your Founding seat runs 10 chats at once, organizations included. It starts as soon as one finishes, in the order shown. |
| Claude Max 20x reached its usage limit. Resumes in 1h 25m. | The subscription the chat runs on hit its limit. It resumes on its own when the limit resets. |
| Waiting for a free agent environment. Next in line. | Every agent environment is busy right now. It starts when one frees up. |
| Too many agent environments started this hour. Retries in 12m. | New agent environments are limited per hour. It retries on its own. |
| 2 messages wait for the current turn to finish. | The agent is still answering. Your queued messages run next. |
| Scheduled. Starts in 3h. | You scheduled it. |
The position and the countdown stay current while you watch. A subscription is named by its provider and plan, never by its email.
Queued messages
Messages you send while a chat works wait in its queue, above the composer. Each shows the model it will run on: the one you chose for it, or the chat's. Messages for the same agent run together, on the latest model they name.
Hover a message, or tap it on a phone, to:
- Edit its text in place. Enter saves, Escape cancels. Attachments stay.
- Move it up or down.
- Remove it.
Clear removes them all. Once the agent has taken the queue, it is running and can no longer change.
Fork or restart
From a chat's menu:
- Fork starts a new chat from this one's first message and a summary of its work, on a new branch from its latest checkpoint. This chat goes on.
- Restart starts the same task again on a fresh agent environment and the same branch, so its pull request carries on. Use it when a chat's agent environment broke. The old chat is stopped and archived.
Either can start from the latest checkpoint or the base branch, on the same or another model, with a note added to the first message. The new chat shows where it came from, and the old one links to it.
From your assistant or scripts
The same is available over the API and to assistants connected over MCP: read_queue, edit_queue and fork_chat.
Settings, API, Queue turns this on or off for your account, in the app and for keys and assistants. It is on by default.