AI Integration Overview
AI Integration lets users ask an AI assistant to find, run, and monitor automations in Automation Hub.
Instead of opening the platform, searching for the right job, filling every input, and checking progress manually, a user can ask for an outcome in plain language. The AI assistant then uses Automation Hub safely through approved tools.
Two Halves of AI Integration
AI Integration has two sides that work together: a place to configure the pieces an agent uses, and a node that runs an agent inside a workflow.
| Half | Where it lives | What it is |
|---|---|---|
| AI Hub configuration | Global Configurator | The reusable building blocks an agent uses — guardrail policies, MCP servers, AI skills, and tool instances |
| AI Agent node | Job workflows | A step that runs an agent — it references the AI Hub objects by id, resolves {{...}} values from the workflow, and returns the result |
You configure the AI Hub once, then point any number of workflows at it. See AI Hub Configuration for the four entities and the AI Agent node for running an agent in a workflow.
The Basic Idea
What MCP Means
MCP stands for Model Context Protocol.
In plain language, MCP is the safe bridge between an AI assistant and Automation Hub. It gives the AI a clear list of actions it is allowed to perform, such as:
- Find available automations
- Read details about one automation
- Start an automation
- Check whether a run is still in progress
- Get the final result
- Pause, resume, or stop a run when allowed
Why This Is Useful
| Without AI integration | With AI integration |
|---|---|
| User searches for the right automation manually | AI helps find the right automation |
| User fills inputs by reading docs or forms | AI asks for missing details in conversation |
| User checks status manually | AI can monitor progress and summarize updates |
| User interprets errors alone | AI can explain what failed and suggest the next step |
What AI Can Do
Find Automations
The AI can search Automation Hub when the user does not know the exact job name.
Example:
"Do we have an automation for database backup?"
Run Automations
The AI can start an automation after it has the required information.
Example:
"Run the March sales report and send it to the finance team."
Monitor Progress
The AI can check if the automation is running, completed, failed, paused, or stopped.
Example:
"Is the report job finished?"
Explain Results
The AI can summarize the final result in normal language.
Example:
"The report was generated and emailed successfully. It processed 1,547 rows."
What AI Should Not Do
AI integration should still follow the same governance rules as the rest of Automation Hub. These are not just conventions — a guardrail policy is a stored, enforced object. The engine turns it into permission rules and resource limits that apply on every tool call and model response.
| Guardrail | Why it matters |
|---|---|
| Use approved automations | AI should run trusted workflows, not invent unsafe actions |
| Ask for missing inputs | AI should not guess important values |
| Confirm risky actions | Destructive operations pause for human approval or are blocked, per the guardrail policy |
| Respect permissions | Users should only access automations they are allowed to use |
| Keep run history | Every execution should remain auditable |
End-to-End Flow
Common Use Cases
| User request | What happens |
|---|---|
| "Send the monthly sales report" | AI finds the report automation, confirms the month if needed, runs it, and reports completion |
| "Check backup status" | AI finds recent backup runs and explains whether they succeeded |
| "Create a user onboarding task" | AI starts the onboarding automation with the required user details |
| "Pause that long-running sync" | AI identifies the run and pauses it if allowed |
| "Why did this job fail?" | AI checks the failed step and summarizes the error |
Section Guide
- AI Hub Configuration - the four configuration entities: guardrail policies, MCP servers, AI skills, and tool instances
- MCP Server - registered MCP servers and how their tools are discovered
- MCP Tools Reference - discovered tools and pre-configured tool instances
- Building AI Agents - guardrail enforcement, human approval, and error handling
- AI Agent Node - running an agent inside a workflow