Test and Debug a Hyperflow

Before publishing, test your Hyperflow to ensure it behaves as expected.

The Aisera Platform allows you to test and interact with your Hyperflows using a live chat interface directly in the Hyperflow Details page. Validate tools, inspect execution details, and ensure that everything works as expected — without affecting end users.

Getting Started

1. In the Aisera Admin UI navigate to AI Automation > Hyperflows and select the Hyperflow you want to test.

2. On the Hyperflow Details page click the Test button.

3. Type a User Query and select a Hyperflow Status — choose Draft to test your latest unpublished changes, or Published to test the live version. If only one version exists, only that option is available.

4. Select OK.

Upon selecting OK a webchat panel appears on the right side of the page, allowing you to interact with the Hyperflow as any user would.

Test in Draft first, then verify in Published before releasing your changes.

How to Test

The test panel supports the full range of Hyperflow behaviors. The examples below show some common interaction patterns you can use to validate your Hyperflow before publishing.

Basic conversation

Interact with your Hyperflow using natural language, just like an end user.

You:    "What's the weather in Seattle?"
Agent:  "The current weather in Seattle is 55°F with partly cloudy skies."

Multi-turn conversations

The test panel maintains full context across turns, enabling natural back-and-forth interactions.

You:    "Check my account balance"
Agent:  "Which account would you like to check — Savings or Checking?"
You:    "Savings"
Agent:  "Your Savings account balance is $12,450.00."

Slot filling and clarification

When additional information is needed, the Hyperflow asks a clarification question. Provide the requested details to continue.

You:    "Create a support ticket."
Agent:  "What is the priority level? (Low, Medium, High, Critical)"
You:    "High."
Agent:  "What is the subject of the ticket?"
You:    "VPN not connecting from home office."
Agent:  "I've created a High priority ticket: 'VPN not connecting from home office'. Ticket ID: INC-4821."

Context switching

You can change your request at any point, even mid-conversation. The Hyperflow detects the shift and adjusts accordingly.

You:    "What's the weather?"
Agent:  "Which city would you like the weather for?"
You:    "Actually, check my account balance instead."
Agent:  "Your account balance is $5,000.00."

Next steps and recommendations

After completing a task, the Hyperflow can suggest next steps or offer additional assistance.

You:    "Reset my password."
Agent:  "Your password has been reset. A temporary password has been sent to your email.
         Would you like to update your password, review recent account activity, or enable
         two-factor authentication?"
You:    "Enable two-factor authentication."
Agent:  "Sure. Would you like to use SMS or an authenticator app?"

Understanding Test Results

The test panel shows the current status of the interaction.

Status	Description
In Progress	The Hyperflow is processing the request or waiting for additional input.
Completed	The Hyperflow has successfully fulfilled the request.
Failed	Something went wrong during execution.

Once the Hyperflow determines your request is fully resolved, the conversation is marked Completed and the input field is disabled. No further messages can be sent in that session, indicating that the agent has finished execution. To start a new conversation, click Stop Test to close the chat, then click Test again.

When testing, the debug panel shows what happened behind the scenes during execution.

Field	Description
Tools Executed	Workflows, Actions, and APIs that were triggered, and their results.
LLM Calls	Details of model interactions. Useful for prompt tuning.
Errors	Any issues encountered during execution.

For additional information troubleshooting information see the Hyperflow FAQ

Tips and Best Practices

Keep the following in mind as you work through your testing.

• Select Stop Test before starting a new session to clear prior conversation context.

• Use actual queries your end users would ask, not just ideal-case examples.

• Test in both Draft and Published mode when making changes to compare behavior.

• If something doesn't work as expected, check the debug panel to see exactly what executed.

• Test with typos, unclear language, and unexpected inputs to surface gaps and edge cases.

Troubleshooting

Issue	What to check
Hyperflow doesn't respond	Confirm the Hyperflow has at least one Action, API, or Workflow configured.
Wrong tool triggered	Review the tool description and system prompt for clarity.
Agent says "out of scope"	The request may not match any configured tool — update tool descriptions or add a new tool.
Clarification question loop	The agent may not understand the slot value — try rephrasing your answer or check the slot configuration.
Different behavior in Draft vs. Published	Expected — Draft tests unpublished changes, Published tests the live version.

FAQ

Does testing affect production users? No. Actions performed during testing do not impact the live Hyperflow or real user conversations.

Can I test as a specific user? [NEEDS VERIFICATION: confirm whether user identity simulation is available in the test panel and whether it retrieves real user data from connected systems.]

Is there a limit to how many tests I can run? No. Each test session is independent and does not affect others.

Why does my Draft Hyperflow behave differently from the Published one? Draft mode reflects your latest unpublished changes. Any modifications to Actions, Workflows, APIs, or prompts are only visible in Draft mode until you publish. Published mode shows the live experience for end users.