Automations
Automations run Agent or Agent Team tasks repeatedly on a schedule in a workspace, and keep the conversation and results of each run.
Intended for: workspace users who need to run tasks on a schedule.
Search keywords: Automation, scheduled task, timed execution, run records.
Create an automation
You can hand off tasks such as daily reports, periodic checks, and repeated analyses to automations. First verify the task instructions, required capabilities, and result format in the target workspace, and then set the schedule. If you still need to discuss the task goal back and forth, start with Conversations.
Create manually
- Switch to the target workspace, go to Automations, and click New Automation > Create manually.
- Select the execution target type and the specific target. An Agent automation is run by one Agent, and an Agent Team automation runs according to the selected team's configuration.
- Enter the task name and the task instructions or task description, set the run time, and configure the context and delivery options described in Section 3 of this page as needed.
- Check whether Trigger once after creation is set as expected, and then finish creating the automation. Return to the list and check the task name, execution target, and enabled status.
You can search the list by task name, Agent name, or team name. After creation, you can edit the task content and run settings, but you can't change the bound Agent or team in the edit dialog. To change the execution target, create a new task and pause the old one as needed.
Create in chat
- Click New Automation > Create in chat, select an available Agent to run the automation, and then click Open new chat.
- In the new chat, fill in the task content, run time, conversation mode, and post-creation status in the prefilled instructions, and check the current Agent and model.
- Send the instructions manually and complete any necessary confirmations. After you receive the creation result, return to the automation list of the current space and check the task and its configuration.
Opening the chat only prefills editable instructions. They aren't sent automatically, and this doesn't mean the automation has been created. The Create in chat entry asks you to select an Agent first. To select an Agent Team through a form, use Create manually.
Set the task and schedule
Task instructions should describe what to do each time, what scope of data to use, and how to deliver the results. For recurring tasks, we recommend specifying a relative time range, such as "Summarize business progress for the previous calendar day", instead of keeping a one-off fixed date in use long-term. For tasks that write to external systems or send messages, clearly state in the instructions the allowed scope of execution and the actions that require confirmation.
Create manually supports four run time types: Hourly, Daily, Weekly, and Monthly. After you select a frequency, enter the corresponding minute, time, weekday, or date, and check the run time preview in the form.
Timezone: Schedules are interpreted in server time and aren't automatically converted to the browser's time zone. The form and details of an Agent automation show Server time zone, and the details also show Next run. Team automations are also scheduled by the server's local time. Before you use them across time zones, confirm the time zone of the deployment environment.
Trigger once after creation is selected by default. If you keep it selected, the task runs one extra time after creation. If you clear it, the task waits for the next scheduled run. If you're told that the trigger after creation failed, first check the created task and the run history in the list, and don't simply create the task again.
To run a task right away, use Run Now (Trigger Now for team automations). A single-Agent automation must be enabled, while the creator of a team automation can still trigger a task manually even when its schedule is paused. When a run of the same task is already in progress, wait for it to finish or terminate it if you have permission. You can't start it again.
Configure context and delivery
Agent automations: model and conversation mode
For the model, you can select Agent default or explicitly specify an available model. When you select Agent default, the model for each run is resolved from the Agent's model configuration. If an explicitly selected model becomes invalid, you need to select another one. A saved task doesn't mean that the model is always available. To maintain execution capabilities, read Agent.
Continuous conversation is off by default, and each run creates a new conversation. When it's on, later runs try to continue in the automation's most recent available conversation, which suits tasks that need to carry on a discussion and its context. If that conversation is no longer available, a new one is created. It doesn't automatically reuse the regular conversation used to configure the task through Create in chat.
When the context of a continuous conversation grows too large, it rotates automatically, while the page history is kept. Don't assume that because the page keeps the history, each run carries all the old content in full. Write key rules and delivery requirements in the task instructions.
Agent Team automations: scope, notifications, and knowledge bases
Team automations use the selected team's configuration, and you can set Analysis Scope in the form as needed. The team form doesn't provide the single-Agent model override or the Continuous conversation switch, so don't assume that these two options are common to all automations.
To receive the results after a run finishes, select an available channel in Channel and complete the corresponding account binding first. For Feishu, Lark, and Slack, you can enter a group chat or channel ID in the form. If you leave it empty, the results are sent based on the task creator's personal binding. Before you select a channel, make sure the recipients are appropriate for viewing the task results.
To keep accumulating results, select Save Result to Knowledge Base, and make sure you have write permission for the target knowledge base. After a run ends, check the results separately where notifications are received and in the knowledge base. The run status, notification delivery, and knowledge base saving are separate things, and a delivery or saving failure doesn't necessarily change the final run status of the task. For maintaining and using knowledge bases, see Knowledge base.
View the run process and results
Open the task details from the automation list and check the configuration and run history. Triggered successfully only means that the run has been accepted, not that the task is complete. Continue to check the run status and the final output.
Agent automations: Run History shows the run time, duration, status, and number of attempts. Waiting, Preparing environment, and Retrying are part of the run process. If a run fails, you can view the stage, error code, and diagnostic details. Click View Conversation to open the corresponding conversation, and use Latest Conversation in the list to quickly view the latest output.
While an automation is running, its conversation input area is read-only to avoid interfering with the current task. Continue the discussion after the run ends, based on your current permissions. If the history shows Linked conversation deleted, the original conversation content can't be restored from that entry. Keeping the run record doesn't mean that the deleted conversation content is kept.
Agent Team automations: In Run History, view statuses such as success, partial success, and failure, and click View Detail on a record to open the details of that team run. To add a reply, confirm something, or troubleshoot a step, handle it in the corresponding run details. Don't conclude from the list summary alone that all the work is done.
To organize run conclusions together with ongoing business goals, you can add progress or link shareable results in Tasks. Running automations on a schedule and organizing results in Tasks are different operations.
Handle errors and maintain tasks
After a failure, first view the run details to tell apart causes such as the model or Agent configuration, permissions and credentials, quotas, and the execution environment. Fix the cause first, and then decide whether to rerun. The task may have already completed some external writes or notifications, so check the actual results before you rerun to avoid duplicate business impact.
Run again: Failed or canceled single-Agent records offer Run again, which requires the task to be enabled with no other run in progress. It creates a new run based on the current task configuration and keeps the original failed record. This is different from the system showing Retrying for the same run. To run a team task again, the creator uses Trigger Now and checks the results in the new run details.
Terminate and pause: The creator of a single-Agent automation can click Terminate this run in the run history. Terminating this run doesn't automatically pause future scheduled runs. For a single Agent, Pause schedule only stops future scheduled runs and doesn't terminate a run that has already started. Clicking Pause on a running team task prompts for a second confirmation, and confirming also terminates the running task. To keep the current run, cancel the operation.
Edit and resume: Use Edit to adjust the instructions and run settings, and reopen the details after saving to check them. Use Enable schedule to resume a single-Agent automation and Resume to resume a team automation. After the schedule resumes, keep an eye on the actual results of later runs.
Delete: When you're sure that the automation is no longer needed, select Delete in the details or the action menu and confirm. There's no way to restore it after deletion, and deleting a running automation also terminates the associated current run. If you only want to stop it temporarily, use Pause. When a team has been deleted, its associated automations can only be deleted. They can't be edited, enabled or paused, or triggered.
Collaboration permissions and FAQ
First confirm the current workspace, and then determine the task's visibility and operation permissions. Agent automations and Agent Team automations currently use different collaboration rules.
| Scenario | View and management scope |
|---|---|
| Agent automations in a personal space | Isolated by creator. You can access only your own tasks. |
| Agent automations in a shared space | Current active space members with access to the relevant Agent can view, edit, enable or pause, trigger, and delete them. The separate Terminate this run operation is available only to the creator. |
| Agent Team automations | Within the currently accessible spaces, personally visible tasks can be viewed only by the creator, and company-visible tasks can be viewed by members of the same company. Only the creator can edit, enable or pause, manually trigger, and delete them. |
Agent automations in a shared space can't use personal Agents. When other members trigger them manually, the actual operator is recorded, but the backend still runs them with the automation creator's identity. The triggering member having permission doesn't mean that the creator's permissions, credentials, and model configuration are necessarily available.
In a shared space, the run conversations of explicitly shared Agent automations are open to current space members. Regular conversations and existing private conversations aren't automatically shared as a result. After a member leaves the space, their original creator identity can't substitute for current membership. Being able to view the run records of a team automation doesn't grant access to other people's private conversations. To adjust space members and roles, read Create a space.
Why didn't it run as expected? Check the space, enabled status, schedule type, and server time zone, and then check the run history to see whether the run was accepted or whether a task is already running. Tasks marked Manual don't run automatically on a schedule.
Why can I see the task but not the action buttons? Team tasks have different view and management permissions, and the single-Agent terminate permission is also narrower than general management permissions. Check the task creator, your current membership, and the related resource permissions.
Why did the run succeed, but I didn't receive a notification or there's no knowledge base result? First make sure that this is a team task with the corresponding delivery options configured, then check the channel binding, the recipient, and the knowledge base write permission, and verify at the target location. Don't rerun the whole business task just because you didn't receive a notification.
- Verify a task or view the results of a run in a conversation first: Conversations.
- Maintain the Agents used by tasks: Agent.
- Find the tools, methods, and knowledge you need: Capability Center.
- View the overall status of the current space: Space overview.
Parent section: Workspaces. You can continue reading directly from the related pages above.