Extract conversation data
Retrieve the full conversation history from an agent chat session as a KNIME table for logging, analysis, or downstream processing.
When this is useful
Extract conversation data when you want to:
- Log agent interactions for auditing or debugging
- Analyze tool usage patterns across conversations
- Feed conversation results into downstream workflow steps
- Store chat histories for later review
If you only need to prompt an agent without capturing the conversation, see Prompt an agent.
Extract conversation data step by step
Prerequisites
- An agent workflow with an Agent Chat Widget (experimental) node
(see Prompt an agent)
1. Configure the re-execution trigger
Open the Agent Chat Widget (experimental) node configuration dialog and locate the Re-execution trigger option.
Set it to Response completed.
This tells the node to re-execute after each completed agent response, making the conversation available at the output port while the chat session is still active.
2. Enable error output (optional)
To include errors in the output table:
- In the Agent Chat Widget (experimental) configuration dialog, open the Error Handling Settings section.
- Check Output errors.
- Set the Error column name (default:
Errors).
When enabled, the output table contains an additional column that reports any errors that occurred during the agent's reasoning or tool execution.
3. Execute and chat with the agent
Execute the Agent Chat Widget (experimental) node and open the chat view.
Send one or more messages to the agent. Each exchange — including user messages, model responses, and tool calls — is recorded.
4. Read the output
After the agent responds, the Agent Chat Widget (experimental) node provides three outputs:
| Output port | Description |
|---|---|
| Conversation | A table with the full conversation history. Each row represents one turn: user messages, AI responses, and tool calls. An optional Errors column is included when Output errors is enabled (see step 2). |
| Tool output data | Data tables produced by tools during the conversation. |
| Combined tools workflow | The assembled workflow containing all tools used by the agent. |
Each row in the Conversation table is tagged with the message role — for example, User, AI, or the name of a tool that was called.
Automatic updates in data apps
When the Agent Chat Widget is used inside a data app, the outputs update implicitly after each completed response without requiring an explicit re-execution trigger.
Result
The conversation is available as a standard KNIME table that you can pass to downstream nodes for filtering, logging, or further analysis.