AI Agent AI Actions

In this article, you'll learn how AI Actions work, how to configure them for your AI Agent, and how to troubleshoot common issues.

Before you start

• You need Owner, Admin, or AI Agent Admin role in Enjo.

• At least one AI Agent must be configured in your workspace. See Create AI Agent.

• The external system you want to connect must be accessible via API, MCP, or a supported integration.

How AI Actions work

AI Actions connect your AI Agent to external systems so it can retrieve data or take actions during a live conversation — for example, checking a ticket status in Jira, unlocking an account in Okta, or creating a GitHub repository.

You configure which actions are available and the agent decides when to invoke them based on the user's request.

💡 AI Actions vs Knowledge Sources — Knowledge sources are indexed content the agent reads from. AI Actions are live operations the agent executes in external systems during a conversation.

When an action is triggered, the agent follows this sequence:

1. Identifies which action is relevant based on the action's description and question variations

2. Asks the user for any missing required information before proceeding

3. Asks the user to confirm before executing the action

4. Runs the action and reports the outcome to the user in natural language

Action types

When you click + Add AI Action, you can create or add an action in the following ways:

Set up an AI Action

Step 1 — Open AI Actions

1. Go to AI Agents and select your agent.

2. Click the AI Actions tab.

3. Click + Add AI Action.

Step 2 — Choose an action type

Select the action type that matches your use case. See Action types above for guidance on which type to use.

Step 3 — Configure the action

1. Enter a clear Name for the action.

2. Write a specific Description of what the action does. The agent reads this to decide when to invoke it — the more specific, the better.

3. Add Question Variations — example phrasings a user might use to trigger this action. Include edge cases and ambiguous phrasings.

4. Configure any required Parameters. Write clear descriptions for each parameter so the agent knows how to extract the right values from the user's message.

5. Set the Relevance Threshold:

   • Higher threshold — the agent only triggers the action when there is a strong match. Reduces accidental invocations.

   • Lower threshold — the agent triggers the action more broadly. Useful for actions that should handle a wide range of phrasings.

6. Toggle Allow Direct Use on if the action should run without asking for user confirmation first. Leave it off for actions that modify data or have side effects.

Step 4 — Save and test

1. Click Save.

2. Go to Bulk Testing to verify the agent invokes the action correctly across a range of test queries.

Action list

Once actions are configured they appear in the AI Actions tab with the following columns:

Troubleshooting

The agent isn't calling the action or is calling the wrong one

Check the action's description and question variations. Add more variations — including ambiguous or edge-case phrasings — to help the agent recognise when this action applies.

The action isn't using parameters correctly

Review the parameter descriptions in the action configuration. The agent relies on these to extract the right values from the user's message.

The action returns an error

Use Postman or a similar tool to test the API call independently. Confirm the endpoint, authentication, and payload are correct before debugging in Enjo.

What's next

• Bulk Testing — Run through user scenarios at scale and verify the agent invokes the right action at the right time.

• AI Flows — Set up multi-step automation when a request requires multiple actions in sequence or conditional logic.

• AI Agent Knowledge — Manage the knowledge sources your AI Agent uses alongside AI Actions.

• AI Agent Settings — Configure behavior, persona, and guardrails for your agent.