Overview#
A workflow is a graph of connected nodes representing conversation stages, with pathways defining agent movement between stages.
Global Node#
Contains instructions applicable across the entire call:
- Company context
- Overall agent goal
- Tone and personality
- Common objection handling
Node-Level Toggles#
Add Global Prompt#
- Recommended to inject Global Node instructions into most nodes
Allow Interruption#
- Default "On": Bot stops and responds to user interruption
- "Off": Bot continues speaking, ignoring user input
Detect Voicemail#
- Available only in Start Node
- Automatically disconnects if voicemail detected
Variable Extraction#
Supported in Agent and End Call Nodes:
- Extract structured variables
- Define name, type, and description
- Examples: customer name, budget, interest
Webhook Node#
- Executed at call end
- Sends payload to specific endpoint
- Use for:
- Storing call data
- Triggering automation
- Syncing with CRMs
Editing Workflows#
- Selectively delete nodes/pathways
- Completely reset workflow
Pathways and Tool Calls#
- Define agent node movement
- Use tool calls to transition between nodes
- "tool" is a reserved keyword for function calls and actions
Saved conversation regressions#
Before replaying, save your workflow and configure an LLM. Replay uses your organization's normal model quota and can incur model usage.
- Open the workflow tester's manual text chat and complete a conversation.
- Select End chat, then expand Saved conversation regressions.
- Enter a scenario name. Optionally add Expected reply contains and comma-separated Expected node IDs. Node assertions use the IDs in the workflow definition, not the display names shown on the canvas.
- Select Save this conversation. The scenario stores the ordered completed user messages and the conversation's starting context. Saving again replaces the scenario for that conversation.
- Select Replay beside a saved scenario. Leave Replay current draft checked to test the saved draft, or uncheck it to test the published version. Save editor changes before replaying. Each replay copies the selected definition at its start, so subsequent edits do not change that replay.
The result shows the run and definition IDs, each assertion, and an expandable transcript. A replay passes when every saved message completes, all expected nodes are visited, and every expected phrase appears in the assistant replies. Phrase matching ignores case. If the agent ends before all messages finish, a check fails. Without optional assertions, the result only checks that all saved messages complete. Model responses can vary between runs.
For example, save a booking conversation with expected reply text appointment
and the ID of your confirmation node. After changing the prompt, replay it to check
that the assistant still reaches confirmation and discusses the appointment.
Scope and limits#
- Save 1–20 completed user messages, each at most 4,000 characters. A replay has a three-minute execution limit; quota failures and execution errors are reported.
- Replay supports conversational Start, Agent, Global, and End nodes, plus Trigger nodes. Workflows containing HTTP/custom tools, MCP tools, call transfers, pre-call fetch, integration nodes, webhook nodes, or QA nodes are rejected before model execution. Use a separate conversation-only workflow and provide required values in its initial context; replay does not fabricate external tool results.
- Regression replay skips completion integrations and webhooks. Ending the original manual conversation still runs its normal completion integrations.
- Text regressions do not verify speech recognition, audio quality, interruptions, telephony, or external delivery. Test those separately with real voice calls and integration checks.
- The list shows up to 100 recently updated saved scenarios. Scenarios are visible only within their workflow's organization and follow the source conversation's retention policy. Delete removes the saved scenario without deleting its source transcript.