Schedule a conversation request

A scheduled Job repeats a request in a conversation using saved settings. It is useful for a periodic summary or inspection that does not need a separately designed graph. Choose Studio when the work needs explicit steps, connected outputs, or folder-triggered processing.

The Runtime must be running when a Job is due. A schedule does not wake your computer or start ABot for you. Missed offline occurrences and failed runs are not retried automatically.

Test the task first

Before scheduling, run the intended request manually with the model and tools you plan to use. Check that it can access its inputs and produces the result you actually want.

For example, a bounded inspection prompt could be:

Inspect the current machine resources using the available system probe.
Summarize the CPU, memory, disk, and uptime values actually returned.
Suggest up to three follow-up checks if the evidence justifies them.
Do not install software, change configuration, or run repair commands.
Do not infer historical trends from this one snapshot.

This is a suggested task prompt, not a built-in command. It requires an available system-probe tool. If the manual request fails, fix the model, tool, or permission issue before turning it into repeated work.

Create the schedule

Open Schedules → New schedule and fill in the form:

1. Give the task a title you will recognize later.
2. Choose the conversation that should receive its runs.
3. Enter the task prompt, including its inputs and boundaries.
4. Choose the model and an available reasoning mode.
5. Review Tool access. This is saved with the Job; do not assume the current chat’s setting will replace it later.
6. Choose the timing and verify the time zone.
7. Select Create schedule and inspect the saved task’s Next run.

For a first check, choose Once a few minutes ahead and keep ABot running. After the recorded run has the expected result, edit the schedule to the recurrence you need. If the model does not expose reasoning controls, the form uses Provider default; enabling controls is a separate model-profile setting.

ABot also has a built-in scheduling capability the model can use. After creating a schedule through conversation, inspect the resulting Job in Schedules. A prose reply saying that something was scheduled is not a replacement for a saved schedule with the correct next run.

Choose the right timing

Form option	Use it for
Once	One request at a specific date and time.
After a delay	A timer relative to when it is saved.
At a fixed interval	A recurring duration. Leave the first run empty to start one interval after saving.
Every day	A daily local time in the schedule’s time zone.
Selected weekdays	A local time on chosen days of the week.
Every month	A chosen day of the month and local time.

The form displays the device’s zone for its exact date/time controls. Daily, weekly, and monthly recurrence uses the separate Schedule time zone field. Check the displayed next occurrence after changing time zones or recurrence settings.

Review permissions before unattended work

The schedule form offers Full and FULL+. Sensitive operations in Full still require approval. FULL+ authorizes task-relevant system actions without additional ABot action prompts, while OS permissions still apply.

Do not enable FULL+ merely to make an error disappear. Prefer a task that needs only the available, appropriate operations. Approved commands run with the Runtime user’s operating-system authority; a working directory is not a shell sandbox. See permissions and projects before scheduling actions that can change your computer or files.

Follow results and stop repetition

Select a Job to see its state, model, tool access, next run, and Run history. Individual run records include their status, saved settings, invocation prompt, and available result or error. Open in conversation takes you to the associated request.

Use Run now when available to test the saved Job, and inspect the result before leaving it active. Pause suspends the schedule; Resume re-enables it. Cancel schedule ends future scheduling. These schedule controls should not be mistaken for stopping an already-running conversation request.

For missing or failed runs, check that the Runtime was running, that the correct environment is selected, and that the saved model and tools are still available. Review the recorded error before manually trying again—especially if a run might already have changed a file or called an external service.