Overview of the Debugger

The debugger in AI Agent Studio helps you to trace, test, and troubleshoot agentic flows in real time. It provides visibility into node execution, context, inputs, outputs, and LLM reasoning, helping you understand the runtime behavior of the agentic flow. Before you publish your workflow or agentic app, you can use the debugger to validate its behavior. It shows how a request flows through the agent, what inputs are passed to each node, what outputs are returned, and how the final response is generated. This visibility helps you isolate defects and reduce trial and error.

Here are some of the tasks you can do using the debugger:

Shorten build cycles: Rerun the agentic flow from a specific node instead of replaying the entire flow.
Improve quality: Catch incorrect routing, malformed payloads, poor tool responses, and weak summaries during design time itself.
Increase reliability: Repeat common scenarios using saved run profiles.
Lower operational risk: Use step-through debugging and output inspection to understand why an agent made a particular decision.
Accelerate root-cause analysis: Trace an incorrect response back to the node where the issue originated.
Improved collaboration: Help product managers, developers, and support teams collaborate by reviewing and discussing node inputs and outputs.

This table provides information about some of the debugging actions supported in AI Agent Studio.

Debugging Action	When to Use	What You Can Inspect or Verify
Run workflow	Run a full test path for the workflow.	Validate the complete workflow.
Run saved profile	Reuse saved test inputs.	Run regression tests for common scenarios.
Inspect nodes	View runtime data for a node.	Inspect inputs, outputs, variables, and context.
Breakpoint	Pause before or at a specific node.	Inspect state of each node before continuing the debug session.
Step through nodes	Run one node at a time.	Verify that each node passes the correct output to the next node.
Resume debugging	Continue the debug session.	Resume workflow run after inspection.
Rerun From Node	Retest the workflow from a specific node.	Retest the selected node and downstream logic.
Pin output	Reuse stable output data.	Verify how cached output affects downstream processing.
Edit output	Test downstream processing with modified output data.	Verify how downstream systems consume the modified payload.
Override node configuration	Test the changes to prompts or configurations.	Verify the effect of temporary changes before applying them permanently.

Here's a screenshot showing the three main working areas in the debugger. working area of debugger

Callout Number	Name	Description
1	Workflow canvas	This canvas displays the workflow nodes and the connections between them. Depending on the workflow, it can include start and end nodes, LLM or agent nodes, Decision nodes, Function or Tool nodes, Loop nodes, Human Approval nodes, and Return nodes. During a workflow run, the canvas shows the runtime status of each node. Completed nodes display a success status, while the active or stopped node is highlighted. Branch labels indicate the path followed after a Switch or If Condition node.
2	Test and Chat pane	In debugger, you can use Ask Oracle to submit a test prompt and review the response. Use realistic prompts that represent the path you want to test, such as a request that must trigger a list, detail, policy, update, approval, or lookup flow. Use the conversation to view debugger status messages that indicate where the run is paused and which node currently has control.
3	Node Details pane	Selecting a node displays the node details that include the Results and Configuration tabs. The Results tab is the primary debugging view, that displays the runtime data for the selected node. It provides the following information: Duration: How long the node took to run. Output: The value produced by the node. Input: The payload passed to the node. Variables: Variables available at that step. Context: Runtime context, prior state, user input, and run metadata. Token metrics: When available for the node type. The Configuration tab enables you to verify the node's setup, such as the LLM prompt, model configuration, bound function, tool configuration, parameter mappings, or routing logic.

Callout Number

Name

Description

Workflow canvas

This canvas displays the workflow nodes and the connections between them. Depending on the workflow, it can include start and end nodes, LLM or agent nodes, Decision nodes, Function or Tool nodes, Loop nodes, Human Approval nodes, and Return nodes. During a workflow run, the canvas shows the runtime status of each node. Completed nodes display a success status, while the active or stopped node is highlighted. Branch labels indicate the path followed after a Switch or If Condition node.

Test and Chat pane

In debugger, you can use Ask Oracle to submit a test prompt and review the response. Use realistic prompts that represent the path you want to test, such as a request that must trigger a list, detail, policy, update, approval, or lookup flow. Use the conversation to view debugger status messages that indicate where the run is paused and which node currently has control.

Node Details pane

Selecting a node displays the node details that include the Results and Configuration tabs.

The Results tab is the primary debugging view, that displays the runtime data for the selected node. It provides the following information:
- Duration: How long the node took to run.
- Output: The value produced by the node.
- Input: The payload passed to the node.
- Variables: Variables available at that step.
- Context: Runtime context, prior state, user input, and run metadata.
- Token metrics: When available for the node type.
The Configuration tab enables you to verify the node's setup, such as the LLM prompt, model configuration, bound function, tool configuration, parameter mappings, or routing logic.