Test and troubleshoot agentic virtual agents

AI-generated summary

Genesys Cloud provides integrated testing and troubleshooting tools for agentic virtual agents through AI Guides, which comprise three core components: AI Studio's preview widget, Architect's bot flow simulator, and Architect's replay mode. The preview widget enables testing of agent roles, guidelines, data actions, knowledge configuration, start/exit behavior, and guardrails. The bot flow simulator allows users to test digital bot flows using the Virtual Agent feature in Architect, with support for Genesys Dialog Engine Bot Flows. Architect's replay mode facilitates detailed analysis and troubleshooting of bot flows connected to agentic virtual agents, providing visibility into tool calls, knowledge lookups, guardrail events, and interaction outcomes to determine why interactions succeeded, failed, or were handed off. Implementation requires Virtual Agent organization enablement and appropriate AI Studio permissions for agentic virtual agents. Flow execution history and replay mode features enable comprehensive agent execution analysis through tool execution review capability, which allows users to examine the sequence of tool calls made during interactions and inspect both data action and knowledge tool calls while monitoring latency and execution status for each tool call. Tool execution statuses include Success, Timeout, Error, and Blocked (guardrail). The interaction outcome feature displays how agents completed interactions—whether Contained, Handed off, or Escalated—along with recorded exit reasons such as tool timeout, customer-requested agent transfer, or guardrail limit exceeded. Interaction behavior analysis functionality enables users to identify the source of unexpected behavior by examining tool execution, knowledge retrieval, guardrails, and agent configuration. These integrated tools work collectively to maintain quality and consistency in customer interactions through comprehensive testing, fine-tuning, and troubleshooting capabilities, supporting optimization of agentic virtual agent performance within the Genesys Cloud platform.

PrerequisitesClick to expand

Test, fine-tune, and troubleshoot your agentic virtual agents to ensure the quality and consistency of the customer interaction experience.

Preview agents in AI Studio

The inbuilt preview widget in AI Studio allows you to test the agent’s role and guidelines, connected data actions, knowledge configuration, start and exit behavior, and guardrails. For more information, see Preview an agentic virtual agent.

Test agents in Architect

Use the built-in Architect bot flow simulator to test your agent with a digital bot flow enabled with Virtual Agent. For more information, see Test your digital bot flow.
To test a bot flow that you create with Genesys Dialog Engine Bot Flows, see Test a Genesys Dialog Engine Bot Flow.

Troubleshoot agents in Architect

Use Architect’s flow execution history and replay mode to investigate how an agentic virtual agent behaved during a conversation.

These tools provide visibility into the agent’s execution, including tool calls, knowledge lookups, guardrail events, and interaction outcomes. They help you determine why an interaction succeeded, failed, or was handed off.

Flow execution historyClick to expand

Replay modeClick to expand

Agent execution details

The flow execution history and replay mode features provide the following agent execution information:

Tool execution

Review tool execution details:

View the sequence of tool calls made during an interaction
Inspect both data action and knowledge tool calls
View the latency and execution status of each tool call
Inspect tool inputs and outputs, if available

The following tool execution statuses are available:

Success
Timeout
Error
Blocked (guardrail)

Interaction outcome

View how the agent completed the interaction, including:

Contained
Handed off
Escalated

You can also view the recorded exit reason to understand why the interaction ended in a particular way.

For example:

Tool timeout
Customer requested an agent
Guardrail limit exceeded

Interaction behaviour

Use the available execution details to determine whether unexpected behaviour originated from:

Tool execution
Knowledge retrieval
Guardrails
Agent configuration

[NEXT] Was this article helpful?

Get user feedback about articles.