The Agent Step allows users to run the existing Egnyte agents as part of a workflow execution. The agent reads the document and understands its content and can automate time-consuming repetitive tasks for users.
AI Actions are available to the customers with the AI add-on enabled. AI Agent step is available to the customers on Elite or Ultimate plans. Please contact the Egnyte account manager or Egnyte Sales team for more information.
Add AI Agent in Workflow Template Step
- Select the AI Agent from the Add Step pop up.
- The system adds the Agent step in a workflow and opens the step configuration panel on the right side of the workflow builder.
- Under the General section, enter a title in the Step name field and add an optional summary in the Description field. Check the Allow edits box under either field to let users modify the step name or description during workflow creation.
-
Select the file(s) to be used as a Source for the Agent.
If any of the selected files are unavailable during the agent step execution, then the step will be marked as failed, and workflow execution will continue based on the Failure Logic configuration for the step.
-
Click Select agent to use on this step field under the Details section then type and search for the specific AI agent to add.
-
Provide custom instructions in the Additional prompt for the Agent field which will be appended directly to the base agent instructions during execution.
- Click the Test AI Agent Execution button to launch the interactive AI Agent execution testing window.
- Select the file to run the test on.
- Confirm the testing document under the Selected file section of the pop-up, or click Select different file to select an alternative sample document. Review or refine the instructions in the Additional prompt for the Agent field's text area, then click the Execute button at the bottom left. Evaluate the generated output on the right side of the screen and click Use prompt to save the verified configuration back to the workflow step.
- Navigate to the Save the output of the prompt execution as a dropdown menu located below the testing button. Users can save the output of agent execution as a Workflow step comment, File comment, File metadata or save it in a New file (.docx).
- If the File metadata option is selected, then users can select any predefined Metadata section. The metadata properties from the selected section are displayed below the Metadata section field.
- If the New file (.docx) option is selected then users can click the Edit button on the blue informational banner to customize the dynamic naming convention (for example, {Document Name}_{Workflow ID}).
- The system opens up a File Saving rules dialog box. Type a custom title in the File name field, or press the opening curly bracket key { to insert dynamic tokens such as {Document Name} or {Workflow ID}.
- Navigate to the Specify the location to save the file section to configure where the generated files will live.
- Select the Same as the original file location radio button for the output to be stored alongside the original source document.
- Select the Specific folder radio button to direct all generated files to a designated directory, making sure the template users have appropriate access permissions to that location.
- Click the Save button in the bottom right corner to finalize the naming conventions and file storage rules.
- Navigate to the Failure Logic section at the bottom of the side panel to configure how the system handles errors during execution of this step.
- Click the dropdown menu underneath the label If this step fails, then to view error-handling options.
- Select Mark step as “Failed” and cancel workflow to mark the Agent step as Failed and cancel the workflow when the agent is not executed successfully within 24 hours.
- Select Mark step as "Failed" and continue workflow execution (default) to mark the step as Failed and workflow execution will continue to the next step or workflow will be marked completed if this is the last step of the workflow when the agent step is not executed successfully within 24 hours.
- Click the dropdown menu underneath the label If this step fails, then to view error-handling options.
- Under the General section, enter a title in the Step name field and add an optional summary in the Description field. Check the Allow edits box under either field to let users modify the step name or description during workflow creation.
Delete the Agent Step
To delete the step, click on the three-dots icon -> Delete either from the right side panel or from the graphic builder.
Connector Authorization
Some AI agents in Egnyte workflows use MCP connectors, such as Gmail, Slack, or Jira. For the successful completion of the agent step execution, the workflow creator must authorize each required connector. The steps mentioned below walks through how authorization works when creating a workflow manually and when created from an automated trigger.
- When creating a workflow manually from a template that includes an agent step requiring connector authorization, the system will display a warning before the workflow is created.
- Click any connector name in the list to open its authorization page in a new browser tab.
- Complete the authorization flow for each connector. Users can authorize all connectors before proceeding, or skip this step and authorize later.
- Click Create workflow to proceed. The workflow will be created regardless of whether authorization has been completed.
- When a created workflow (manually or automated) reaches the agent step, the workflow creator receives an email notification.
-
Click Authenticate in the email to open the workflow and view the agent step.
- In the agent step, review the list of connectors requiring authentication.
- Click any connector name in the list to open its authorization page in a new browser tab.
- Click each connector link in the error message to complete authorization.
- Once all connectors are authorized, click Retry to execute the agent step.
The step is marked as failed if connectors are not authenticated within 24 hours of the step starting. If any tool within a connector requires user permissions, those specific tools are blocked in the workflow.