The Chatbot traces menu makes it possible for ADMIN and BEHAVIOR_DATA_USER users to see the content of conversations between users and Chatbot components.
Selecting Open details opens a specific conversation. It is then possible rate the conversation or to select individual messages from the Chatbot to see the following information:
| Information | Description |
|---|---|
| Metadata | Shows the latency, the time from the LLM call to the beginning of the response (TTFT), the number of input tokens billed by the provider, and the number of output tokens. |
| Summary | Appears after the first Chatbot message. Chatbot components reformulate user queries to create more efficient semantic search queries. Reformulated queries take into account past user messages to create a standalone semantic search query. |
| Semantic search results | The results of a semantic search using the user's query, or the reformulated user query (also known as "summary"). |
| Chatbot instructions | The prompt configured in the AI profile used by the Chatbot component for the conversation. |
Chatbot feedback must be enabled in the Chatbot component.
Sort the Chatbot traces table by User rating to identify unsatisfactory conversations. Unsatisfactory conversations can reveal documentation gaps, or the need for a better prompt in the AI profile.
- As a beta release, the Chatbot traces menu benefits from ongoing improvements.
- The Traces menu does not show conversations with the Agentic chatbot.
- By default, the Chatbot traces menu shows the last 50 conversations. It is possible to see up to 100 conversations by using the Pick a period filter.
- It is possible to access Chatbot traces programmatically using the Search Chatbot traces or Get a Chatbot trace details web service.