Skip to main content
At the end of this page, you can run your current flow in Test Mode, trace every event, and clear validation problems before publishing.

​
Before you start

Before you start, sign in to the dashboard and select the organization you want to work in. Test runs never affect your live agent, and test conversations never count toward billable conversation usage.

​
Run a test conversation

Tool calls in a test run can perform real actions in connected services, such as sending an email or creating a record. See What test runs simulate for the exceptions.
1

Select Test in the toolbar

The Test button carries the tooltip “Test agent”. It opens the Test Mode panel on the right side of the builder.
Test Mode panel open beside the flow, with a running chat, the Run, Pause, and Reset controls, and the Variables section showing the System group with the conversation transcript
The right panel shows one thing at a time. Opening Test Mode replaces the gambit editor or the versions panel.
2

Select Run

Before the first run, the panel shows an empty state. It reads “Test execution is not yet active” and “Press Run to start the conversation flow”.Select Run to start the conversation flow.
3

Answer as an end user would

Reply to each gambit the way a real end user would. Watch the Variables section fill as the flow captures values.
4

Pause when you want to inspect a step

Select Pause to stop the run in place. The button becomes Resume, which continues from the same point.
5

Reset to start over

Select Reset to clear the run. Then select Run to test the flow again from the top.
A test run uses the flow you are editing. While a draft is open, the run uses that draft, and switching drafts stops the run. See Drafts and publishing for the environment chip and drafts.

​
The Test Mode panel

The panel header reads Test Mode and holds the run controls plus a close X. During a run, a Test user card above Variables names the test end user. Open user → opens that end user’s profile in the test environment.

​
Variables

The Variables section lists every value the run has captured, grouped by scope, with a count in the header. Before anything is captured, it reads “No variables set yet”. Select the pencil icon next to a response or flow variable to edit its value in place. Press Enter to apply the override. An override lets you test a branch without replaying the whole conversation to reach it. See Use variables. Select Inspect transcript to add the conversation so far to the System group as system.transcript. Select Hide transcript to remove it from the list. See Insert the conversation transcript.

​
Event Log

The Event Log is a collapsible section that records what the run did at every step. Before the first event, it reads “No events yet”. Filter chips carry a count per category: All, Lifecycle, Flow, AI, Routing, and Errors. Start with Errors when a run goes wrong. Then switch to Routing to see which branch the flow took. Each row expands to show its detail. A category with no events keeps its chip at zero rather than hiding it.

​
What test runs simulate

The Meeting Scheduler and verification codes use simulated results in test conversations.
  • The Meeting Scheduler creates no calendar event, invite, or email in a test conversation. Google Calendar and Outlook tests keep your configured working hours, timezone, duration, and minimum notice while simulating calendar availability. Simulated CodeKit integrations use weekday 9:00–17:00 working hours instead. See Book meetings with the Meeting Scheduler.
  • A verification code is not sent. Enter 000000 to pass a Verification Code component, or a text question set to verify by one-time code.
  • Tool calls from a test run carry a Test tag in the toolkit’s Logs. See Monitor tool calls.

​
Validate the flow

Validation catches structural problems that a test run might not reach, such as an unreachable gambit or a duplicate variable name.
1

Open the Problems panel

Select Validate at the top right of the builder. Validation runs and the Problems panel opens. Toggle Validation Panel in the command palette, and Cmd+Shift+V or Ctrl+Shift+V, do the same.
Problems panel at the bottom of the builder listing warnings such as unreachable gambits, each with a Go to link
2

Read the counts in the Problems header

The header shows the error and warning counts, or “No issues”. A “Last checked” timestamp shows when validation last ran.
3

Select Go to on a problem row

Problems group under Errors and Warnings. Each row has a Go to link that focuses the affected gambit in the builder.
4

Fix the gambit, then re-run validation

Select the refresh control, with the tooltip “Re-run validation”, after each fix. A clean flow shows “Flow validation passed. No issues found.”
Errors block publishing, so clear them before you publish. The publish dialog shows the same result in its Pre-publish validation panel. See Drafts and publishing.

​
Control the debug detail end users see

Open your agent, select Configure in the builder sidebar, then select Debug in the Settings list on the Configure page. It sets how much AI detail the widget shows during a conversation.
Debug selected in the Settings list on the Configure page, with switches for AI reasoning, tool calls, and knowledge citations
Select Save Debug Settings to apply. These switches change only the widget. In the dashboard you always see the full reasoning and tool detail for every conversation. See What end users see.

​
Review test runs in Activity

Every test run is stored as a conversation in the test environment. Open the Activity tab and switch the source picker to Test · Published flow or to your draft’s entry. The timeline shows every message, AI turn, and tool call from the run. See Watch conversations.

​
Verify

Run the flow and send an answer that should take a specific branch. The Routing chip in the Event Log shows the branch decision, and Variables shows the captured value. The Problems panel reports “No issues”.