Every time your agent responds to a message, there is often more happening behind the scenes than the reply itself. The agent might search your knowledge base, look up a customer record, send an email, or delegate a task to another agent. The activity log gives you full visibility into all of these actions.
In the chat view, the activity appears in the Activity rail alongside the conversation on wider screens. The rail has three tabs:
While the agent is working, a live indicator shows in the rail header and steps stream in as they happen. Each message keeps its own step timeline, so you can review it again after refreshing.
You can also find activity in:
Each entry in the activity log represents a single action the agent took. Entries appear in chronological order (top to bottom) and include:
When the agent uses a tool — such as searching a knowledge base, sending an email, or creating a job in your management system — the activity log shows:
When the agent searches your knowledge base, the activity log shows the exact query it used and the documents that were returned. This is useful for debugging situations where the agent gives an incorrect answer — you can see whether it found the right documents or whether the search query missed the mark.
If the agent hands off a task to another agent (in an agent team), the activity log shows:
This makes it easy to trace exactly how a multi-agent conversation played out.
When the agent interacts with an integration — for example, fetching data from simPRO or creating a calendar event in Google — the activity log records the request and response. Sensitive data like authentication tokens is redacted, but you can see the structure of what was sent and received.
Each tool call in the activity log shows its status:
| Indicator | Meaning |
|---|---|
| Running | The tool call is in progress. You will see this during real-time streaming. |
| Done | The tool call completed successfully. |
| Failed | The tool call failed. Expand the entry to see the error message and details. |
Steps that run as part of a workflow can also show a Pending status while they are queued and waiting to start.
The activity log is your primary tool for understanding why an agent behaved a certain way. Here are common debugging scenarios:
Agent gave wrong information
Check the knowledge search entries. Did the agent search for the right terms? Did the correct document come back in the results? If not, you may need to update your knowledge base or refine the agent’s persona to guide its search behaviour.
Agent did not send an email
Look for a “Send Email” tool call. If it is missing, the agent may not have recognised the intent. If it is there with an error status, the email integration may be disconnected or misconfigured.
Agent took too long to respond
Check the duration on each tool call. A slow response usually means one tool call took a long time — often a large knowledge base search or an external API that was slow to respond.
Agent gave a generic answer
If there are no tool calls at all, the agent answered from its general knowledge instead of searching your data. Check that the agent has the right tools enabled and that your knowledge base is connected.
The Activity rail header shows the usage for the latest turn, which is useful for understanding cost and performance:
This gives you a per-turn read on how much work each response involved, without reading every step in detail.
