Start (Realtime)

The Realtime Start Node operates differently from a traditional text chat node. Because this canvas is built for ultra-fast, continuous audio loops (like SIP Call Centers or headless Kiosks), the flow is almost never triggered by a manual "send" button.

Instead, the Realtime Start Node natively listens for system-level API triggers.

[!IMPORTANT] The Realtime Flow must be initiated using the Async Worker endpoint: POST /api/v1/openai-realtime/{FLOW_ID}/queue/enqueue.

How It Works

Instead of waiting for typed chat, the Realtime Start node receives a structured JSON form payload from the voice/realtime client or from an external system (for example CRM) that enqueues the call.

When the remote queue event fires, Lyntaris automatically pushes:

1. Session Context: The UUID matching the active phone line or physical Kiosk.
2. State arrays: Context from your deployment configuration, such as active_agent_id or current_media_prompts.
3. Heartbeat Triggers: A continuous ping sequence (typically every 5 seconds) begins arriving at the Start node, letting the Flowise engine know the caller is still active, safely resetting the 60-second execution timeout.

Best Practices

• Do NOT assume standard graphical UI loading delays. When the Start Node fires, you must be ready to stream the LLM response instantly, or the end-user will hear dead silence on the line.
• The question field on the payload acts as the caller's spoken transcript, processed instantly through the Node into the downstream flow.