Candela tracks every conversation as a trace, giving you visibility into what happened, when, and why. You can retrieve and review the full event log for any conversation to help you:
Review what Candela did on behalf of yourself or another user
Diagnose unexpected behaviour
Satisfy audit requirements
How conversation tracking works
Each Candela conversation is recorded as a trace. A trace captures the complete sequence of events in a conversation, including every:
Message you send
Tool the agent calls
Result the agent receives
Agent transition that occurs
Confirming write operations
Candela distinguishes between read operations, write operations, and destructive operations.
Before executing a destructive action such as deleting a file from LUSID Drive, Candela always asks you to confirm. This confirmation step is recorded in the trace alongside the your response, providing a clear audit record of user intent.
Viewing traces
Provided you have administrator privileges, you can view traces via the web app or retrieve them programmatically via the Candela API.
Navigate to Insights > AI Assistant Traces to view all conversation traces for a given time period, and retrieve the trace ID for a particular conversation:
Viewing the trace for a particular conversation
Select the View icon on any trace to open the full details of a conversation:
Each trace includes:
The complete event log, showing every message, tool call, tool result, and agent transition
A diagram of the agent workflow for the conversation
A search bar for navigating the event log
A link to the correlated request logs
Retrieving traces via API
You can also retrieve traces programmatically via the CandelaTraces API, which is useful for building audit reports or integrating trace data into your external tools.
For example, to list all traces:
curl -X GET "https://<your-domain>.lusid.com/insights/api/candelatraces"
-H "Authorization: Bearer <your-API-access-token>"