Set up an Automation
Ask an Agent in the Room to describe the process before saving it. Specify the information to read, the actions to take, when to run, and where to send the result. For example:Every weekday at 9 a.m. Singapore time, summarize open P0 issues from our incident App in this Room. Include an owner and link for each issue. Do not change the records.Review the proposed process and, when practical, a sample result. The Agent can save instructions for judgment-based work or TypeScript for tasks that need batching, pagination, or recovery. The Automation runs separately from your ordinary conversation, using its saved definition and authorized resources.
Choose a trigger
Record triggers can filter on field values, such as
priority being P0. A filter on the current value is not a test that the value just changed; see the record-change tutorial.
Schedules use the Automation’s timezone. The production minimum interval is normally 10 minutes. Supported Connection delivery flows can also create Automations for incoming service events; connecting an account alone does not enable delivery.
Review results
Results appear in the Room’s main conversation by default. You can instead keep them in the authoring thread, start an Agent thread from the result, or keep them only in run history. Open the Automation to inspect its process, prompts, trigger, output, files, and credit use. Run history distinguishes completed, failed, and skipped runs. A skipped run did not do the work. For an external Agent, a handoff status alone does not establish completion; review its output and any resulting changes. Run now starts a real execution, including its normal sends, writes, and credit usage. Saving or resuming an Automation does not run it immediately.Pause and recover
Pause stops future automatic runs. Resume re-enables the trigger. Repeated failures or credit protection can also pause an Automation; inspect the reason and resolve it before resuming. Archiving the Room pauses its Automations. Definition changes are saved as revisions. A rollback selects an earlier definition for future runs, while a run already in progress keeps its original revision.Receive webhook events
Treat the webhook URL like a password. Configure a supported signing profile when available so Kylon can verify the sender’s signature. Copying or revealing the URL keeps it working; rotating it invalidates the old URL, and revoking it stops the endpoint. Accepted deliveries arriving while paused are held for up to seven days, with a limit of 1,000 per endpoint. Resuming does not replay them. After resuming, the owner can choose to replay the held events or discard them. Replay performs the normal actions of a run.Automations managed by an App
An App can include Automations in its registration or source. Managed by App means the App owns the definition: change or remove it through the App. You can still pause, resume, run it manually, and adjust credit protection from the Automation controls.Control spending
Set daily, weekly, or monthly budgets to warn or pause when the Automation exceeds its budget. Workspace limits for API calls made by code are separate. See Billing & credits.Developer reference
The CLI command group is namedworkflow. Its reference covers trigger JSON, timezone and delivery flags, revisions, webhook controls, and the @kylon/workflow-sdk. Code that needs information across runs should use Automation state storage rather than local scratch files.