Purpose
This article explains how to inspect the streaming events generated by the orchestrator by using the Kafka interface.
Overview
The platform uses Kafka as the event backbone for online orchestration. As the agentic system runs, the orchestrator emits events that describe the execution flow.
These events can be inspected in the Kafka interface, which makes it possible to analyze how the system behaved at runtime. This is one of the main ways to get full visibility into what happened across agents and tools.
When to use Kafka inspection
Inspect Kafka events when you want to:
- understand the full execution path of an agentic request
- review the sequence of internal actions
- validate that events are being emitted correctly
- debug orchestration behavior beyond what is shown in chat
Kafka is especially useful when deeper operational visibility is needed.
Step 1: Open the Kafka interface
To inspect streaming events:
- open the Kafka UI
- navigate to the Topics section
This is where the available Kafka topics can be explored.
Step 2: Open the orchestrator events topic
Inside the Topics section, find and open:
orchestrator-events
This topic contains the messages emitted by the orchestrator during agentic execution.
[Insert screenshot here: Topics page with orchestrator-events highlighted]
Step 3: Inspect the messages
Once inside the orchestrator-events topic, you can inspect the emitted messages.
These messages provide visibility into the execution flow and can be used to understand what happened inside the system while requests were being processed.
What you can learn from Kafka events
Kafka messages can help reveal:
- which actions were emitted by the orchestrator
- how execution progressed over time
- how different parts of the runtime system interacted
- whether expected events were published during a request lifecycle
This gives a broader and more complete view than the final chat response alone.
Kafka versus chat visibility
The chat interface provides a user-facing view of agentic activity. Kafka provides a lower-level event stream.
A useful way to think about the difference is:
- Chat shows a curated runtime view for interaction
- Kafka shows the streaming event backbone behind that runtime
When deeper debugging or validation is required, Kafka is often the more complete source.
Typical workflow for event inspection
A common workflow is:
- deploy the agentic system
- interact with it through the chat interface
- open Kafka
- navigate to
orchestrator-events - inspect the emitted messages corresponding to the interaction
This makes it possible to compare what the user saw with what the orchestrator actually emitted.
Common use cases
Verifying that orchestration events are emitted
If the chat interface or another consumer is not showing expected behavior, Kafka is the first place to check whether the events themselves were produced.
Debugging agentic flows
Kafka is useful when a flow behaves unexpectedly and a deeper step-by-step view of the execution is required.
Understanding the full runtime trace
Some details may not be surfaced directly in the UI. Kafka provides a more complete record of the event stream behind the agentic system.
Common issues
The topic exists, but no messages appear
This may indicate that the system is not currently receiving requests, that the deployment is incomplete, or that event emission is not working as expected.
Messages appear, but they are hard to interpret
In that case, compare the event stream with the online YAML configuration and the observed chat behavior. Kafka is most useful when read together with the system definition.
Chat shows activity, but Kafka looks incomplete
This may indicate timing, filtering, or consumer-side differences. The orchestrator event topic is still the best place to confirm whether events were published.
Best practices
Use Kafka together with chat
Chat is good for interactive observation, while Kafka is better for deeper runtime inspection.
Inspect events after a known request
It is much easier to interpret Kafka messages when you know which interaction triggered them.
Keep the orchestration configuration nearby
Kafka events are much easier to understand when you can compare them directly with the service graph and prompts defined in the online YAML.
Comments
0 comments
Article is closed for comments.