Automations
Scheduled and event-driven AI workflows
Automations allow you to run Operator workflows automatically—triggered by external events via webhooks or executed on a schedule. Each automation is a reusable definition that specifies what the AI should do, which capabilities it can use, and what context it needs.
Access: Automations require Power User or higher access (level 30+).
When triggered, automations create a dedicated agent conversation that runs in the background, persisting results and handling destructive operations with confirmation flows.
How Automations Work
An automation consists of the following components:
- Name & Description — Identify the automation in the UI and run history
- User Instruction — The natural language prompt that tells Operator what to do
- Trigger Type — How the automation is initiated (webhook, schedule, event, or manual)
- Agent — The system-defined AI persona to use (e.g., System Architect, Business Specialist)
- Capabilities — The specific capabilities enabled for this automation
- Instance Context — Optional ERP.net instance and enterprise company context
System Agents
Each automation uses a system agent that defines its personality and default capabilities:
| Agent | Description |
|---|---|
| Client Advisor | Customer-facing agent for commercial self-service: orders, invoices, and product information |
Background Runs
When an automation runs, it creates an agent conversation that processes in the background. You don't need to keep the browser open—the conversation continues server-side until completion. Active conversations appear in the Active section of your conversation list with real-time status updates.
Destructive Operations
Automations follow the same safety rules as interactive chat. When a destructive tool (creating records, modifying data) is about to execute, the automation pauses and waits for confirmation. You can approve or reject the action from the conversation view.
Trigger Types
- Webhooks — Trigger automations from external systems via HTTP
- Schedules — Run automations on a recurring schedule
- Events — React to events from connected systems
Run Status
Each automation run is tracked with the following statuses:
| Status | Description |
|---|---|
| Running | The automation is actively executing |
| Completed | The automation finished successfully |
| Awaiting Confirmation | A destructive operation requires user approval |
| Waiting | A follow-up event arrived while the execution was busy; it will be processed automatically when the current run completes |
| Failed | The automation encountered an error |
Follow-Up Continuity
When an automation is triggered by events that belong to a thread (e.g., email replies in the same conversation), follow-up events can continue within the same conversation instead of creating a new one. This preserves the full conversational context, allowing the AI to reference earlier messages and decisions.
How It Works
- Thread Detection — Events carry a
thread_id(e.g., an email conversation ID). When a new event arrives, the system checks if a conversation already exists for this automation + thread combination. - Idle Conversation — If the existing conversation is idle, the follow-up message is appended directly and the agent resumes processing.
- Busy Conversation — If the existing conversation is still running (or awaiting confirmation), the follow-up is queued as a Waiting run. Once the current run completes, the system automatically drains waiting runs in order.
- New Thread — If no prior conversation exists for this thread, a new conversation is created as usual (with the thread ID stored for future follow-ups).
Configuration
Follow-up behavior is controlled in the Follow Up tab of the automation editor:
- Continue in same conversation — When enabled (default), follow-up events reuse the existing conversation thread
- Custom follow-up instructions — Optionally provide different instructions for follow-up events (e.g., "The customer has replied. Review their response and continue the conversation.")
Capabilities Per Automation
Each automation has its own set of enabled capabilities, independent of your user settings. This allows you to create focused automations that only have access to the tools they need.
When an automation uses a user-defined agent, the automation's capabilities are merged with the agent's own capabilities. This means you can grant additional capabilities (such as Email) to the automation even if the agent doesn't include them by default. The agent's capabilities are always available, and the automation can extend them further.
For example, a "Create Daily Report" automation might only need:
- Query ERP Data — To fetch the data for the report
- Document Management — To generate a document
Restricting capabilities follows the principle of least privilege and prevents automations from performing unintended actions.
Run History
Every automation run is logged in the run history, capturing:
- Trigger type (webhook, schedule, event, or manual)
- Start and completion timestamps
- Event/webhook payload (if applicable)
- Final status and any error messages
- Link to the full conversation transcript
Access run history from the Automations page to monitor runs, debug issues, and review what actions were taken.
Creating an Automation
- Navigate to the Automations page from the menu
- Click Create Automation
- Fill in the name, description, and user instruction
- Select the trigger type
- Choose which agent persona to use
- Configure the capabilities the automation can use
- Optionally configure instance and enterprise context
- Save the automation