ZeroLog
Private preview
EN DE

Product

Commands that keep running while the chat moves on

In a chat with a workspace, the model can start a long command in the background, keep working, and pick up the result when the command ends, even if nobody has the chat open.

User impact: Targeted change

People who use a chat with a workspace: the model can run long commands in the background, watch for an event, set a wake-up or propose a recurring instruction you confirm; the workspace rail shows and stops them. Each wake-up turn is charged like a normal turn. No action needed.

A background command returns at once. When it ends, the chat gets a new turn with the exit code, the end of the output and the file path.
Text version

Start in background: the model gets a task id at once. Model keeps working: the output goes to a file in the workspace. Command ends: the chat gets a new turn with the result.

A build, a test run or a download no longer holds the chat. The model starts the command in the background, gets a task id and keeps working. The output goes to a file in the workspace. When the command ends, the chat gets a new turn with the exit code and the end of the output, even if nobody has it open.

A watcher waits for an event instead: each new line of output wakes the chat, at most every 30 seconds, until a time limit the model sets.

The model can also set a one-time wake-up for the chat or propose a recurring instruction. A recurring instruction runs only after you confirm it, with a fixed number of runs, a spending cap and an end date.

The workspace rail of the chat shows what runs, its output, and buttons to stop a task or confirm a proposal.

A background command has the same rights as the same command in the foreground, and limits apply. Each wake-up turn is charged like any other turn and does not start without credit. Details are on the Jobs and Automation pages.

No action is needed.

Back to Newsroom