Skip to content
Ankole

Automation blueprints

For AI Agents: the Markdown version of this page is at https://ankole.agentbull.com/en-US/docs/automation-blueprints/index.md. The documentation index is at https://ankole.agentbull.com/en-US/llms.txt.

Choose a trigger and a consumer separately. A schedule, Checkback, raw callback endpoint, or Connector event supplies a CloudEvents input. An Agent handles judgment and conversation; an Applet Action handles deterministic work. Bindings can deliver one source to several consumers.

Common patterns

Goal Configuration Result
Daily briefing Agent schedule with a task and delivery destination The Agent gathers information and sends a summary.
Frequent mechanical check Applet Job with an owned schedule The Action checks conditions without a model turn and records the result.
One-time follow-up Agent Checkback The originating conversation resumes at the requested time.
External change Connector subscription or a raw callback endpoint A binding starts the selected Job or Agent.
Extended research Agent schedule followed by a Background Agent Job Research runs independently and reports back to the owning conversation.
Quiet channel monitoring A quiet chat binding plus an Agent schedule Messages accumulate without replies; the schedule produces a recap.

Build a deterministic monitor

Publish the checking logic as an Applet Action. Create a Job in the Applet’s Automation panel, set its configuration, and add a schedule or external source. Use ctx.signal.data for event data and ctx.state for durable cursors or the last observed value. Call platform tools through AnkoleRuntime.mcp and choose an explicit connection when several exist.

Test a normal condition, a changed condition, and an upstream failure. Inspect actual Run inputs, results, and logs. If a change needs human attention, implement the notification explicitly through the relevant platform tool; a successful silent Run does not send a chat message automatically.

Choose the right kind of work

Use a direct Agent consumer when every event needs memory, interpretation, or a written response. Use an Applet Job when code can make the decision. A Workflow provides bounded parallel subagent judgments inside one Agent turn. A Background Agent Job handles longer Agent work; it is a separate execution model from an Applet Action.

See Automation Jobs for setup and recovery, Schedules for timing, and Webhook delegations for callbacks.