AI

Schedule agent work

Schedule Codex prompts or Bash commands, review runs, and pause automation.

Start with Open Codex chat if you have not used the agent before.

Schedule Codex work in a thread

Open a Codex chat's thread menu and choose Automation settings…. The Thread automation dialog lets you configure recurring work in that thread. Availability and active-automation limits depend on the project and membership tier; an admission limit can prevent enabling another schedule.

  1. Enable Enable scheduled automation for this thread, give it a Title, and choose Codex prompt as the Run type.
  2. Enter what Codex should do on each scheduled run.
  3. Under Schedule, choose Daily or Every N minutes and select the weekdays under Repeat on.
  4. For Daily, enter Run at (24h). For an interval, enter Repeat every (minutes) and either enable Run all day or set From (24h) and Until (24h). Until must be later than From on the same day; this form does not accept a window crossing midnight.
  5. Check Timezone, choose the Pause after unacknowledged runs limit, and click Save.

Use a named timezone such as Europe/Madrid. The form initially uses your browser's timezone, so verify it before saving a schedule intended for another location. After saving, inspect the thread's schedule summary and Next run display to confirm the intended timing.

Run a Bash command on a schedule

For a repeatable script that does not need a new agent decision each time, use a command automation. Availability and active-automation limits depend on the project and membership tier. Commands run unattended and can modify project data; begin with a read-only or idempotent command you have reviewed.

  1. Open a Codex chat's Automation settings…, enable the schedule, enter a Title, and select Bash command under Run type.
  2. Enter the Command, for example pwd for a harmless first run.
  3. Set Working directory explicitly for commands using relative paths. Otherwise it is derived from the chat file path, which may be a generated chat directory; if no parent is available, it falls back to / in the project runtime.
  4. Set Timeout (seconds) and Max output to capture (KB) to fit the command, then configure the schedule and click Save.

The run report records the command and working directory, an exit code or signal when available, and captured stdout and stderr. A truncation notice means output exceeded the capture limit; it does not mean the command's entire output was retained in chat. A command with no captured output is reported as such.

Use Run now to try the configured command before relying on scheduled runs, and inspect its result. That manual run leaves the next scheduled occurrence in place, so account for both when the command changes data.

Control and acknowledge scheduled runs

Open the automation details in its chat thread to see the schedule, latest result, and available controls.

Overlapping runs are not queued. Check Last run, the status, and any displayed error when investigating a missing or unsuccessful result.

Finished automated runs increase the unacknowledged count, including successful, failed, or interrupted runs and Run now executions. At Pause after unacknowledged runs, the automation pauses until you review it. Click the unacknowledged-count button to clear the count; sending a new request to Codex in that thread also clears it. Acknowledging does not itself mean resuming: use Resume when you want a paused schedule to continue, and confirm the next-run display afterward.