Before you start
Before you start, sign in to the dashboard and select the organization you want to work in.- You need an agent with at least one conversation. Runs from the builder’s test panel count.
Two places to watch a conversation
The dashboard shows conversations in two places. This page covers the agent Activity tab.
For the inbox, see Work the inbox and Agent Workspace.
Open the Activity tab
1
Open your agent
In the dashboard, select AI Agents, then select your agent.
2
Select Activity in the builder sidebar
The header shows the agent name with the sub-label Activity.
3
Check the view toggle
Timeline is selected first. Data view sits beside it and shows one card per conversation.
4
Select a conversation
The most recently active conversation opens on its own. Select any other row to open its timeline beside the list.
Choose the source: live, test, or a draft
The source picker sits beside the date range. A green dot marks the Live source, and an amber dot marks any test source.
When no drafts exist, the draft section reads No drafts yet. Changing the source clears the selected conversation. Test and draft conversations are never billed. See Conversation usage.
Pick a date range
The date range picker defaults to the current month. Open it to pick presets such as Today, This week, Last 30 days, or Last 3 months, or type a number of days up to today. Select Apply to filter, or Reset to return to the current month. The conversation list and Data view both follow the selected range.Find a conversation in the list
The list on the left covers the selected range and source. Type in Search conversations… to filter the loaded rows by display name. Each row shows an initials avatar, the end user’s name, a channel icon with the channel name, a status dot with the status, and a short date. An end user with no name appears as Visitor #a1b2c3. Statuses are active, pending, snoozed, resolved, and abandoned. The list loads more rows as you scroll. When the range holds nothing, the list reads No live conversations yet or No test conversations yet.Read the timeline
The conversation header shows the avatar, the end user’s name, and chips for channel, environment, and status. A meta line below it gives the start time, message count, and event count. Filter chips above the feed narrow the timeline: All, Lifecycle, Flow, AI, Routing, Errors, and Messages. When a chip matches nothing, the feed reads Nothing matches this filter. The feed interleaves message bubbles with event rows, sorted by time and split by day separators. Each event row shows its time, a category dot, an icon, and the event title. Images sent in the conversation open in a full-screen viewer.Expand an event
Select a row to expand a detail table of its fields. A Reached row lists the gambit, its type, and the gambit it came from. A Captured row lists the input type, the variable, and the value. Error events show in red with an error pill. A Continued in a new conversation row means a campaign reply replaced this conversation with a new one. Its detail table shows Started by as Campaign reply. Previously handled by shows what controlled the conversation before the reply. See Handle campaign replies. A Message delivery failed row means a message did not reach the end user on WhatsApp. The Error field gives the reason. When the failed message was the question of a gambit, the reason also states that the flow is waiting for a reply. For example, a question with no text is not sent, so the end user cannot answer it.Read an API call
A gambit API action expands into an API call detail card. The card header shows the method, the URL, the status code, and the duration. Tabs switch between Response, Request, and Headers, and a footer names the variables the call captured. Secret headers are masked, and long bodies are trimmed at capture.Inspect an AI turn
Each AI reply in the feed carries a thought pill above the bubble. Select it to expand the reasoning and the tool calls for that turn. Select the AI chip to isolate the AI events around that reply. The AI agent responded row lists the model and the tool call count, and a Tool row reports whether the call succeeded. In the dashboard you always see the full reasoning and tool detail. What end users see of it is a separate setting. See What end users see.Search captured variables
1
Open the variables panel
Select the braces icon in the conversation header. The Captured variables panel opens on the right.
2
Search by name
Type in Filter variables… to narrow the list. The filter also matches the gambit that captured the variable.
3
Browse the groups
Variables sit in collapsible groups such as response, api, flow, enduser, utm, and url. Each group shows its count, and an empty group reads empty.
4
Jump to the capture point
Select a variable name with the tooltip Jump to where this was captured. The feed scrolls to the capture event and flashes it.
Read the Data view
Select Data view in the header to see one card per conversation instead of the timeline. Cards follow the same source and date range. When nothing matches, the view reads No conversations in this range. Each card opens with the avatar, the end user’s name, the start time, and the status. Two sections follow:
Handover moments appear as their own rows in Gambit data, for example:
- Alex Morgan took over the conversation
- AI agent took over the conversation
- Returned to the flow
- Resolved by the end user