Condition Agent (Realtime)

Unlike a standard Condition Node which relies on deterministic JavaScript logic ($revenue > 500), a Condition Agent employs a generative LLM model to determine true/false routing visually on the canvas.

Use Cases in Live Audio

In a realtime SIP or Kiosk deployment, callers rarely speak in structured JSON. If a Call Center worker asks: "Would you prefer a morning or an afternoon appointment?", the caller might say "Well the kids get out at 3, so probably before that."

A traditional JavaScript Condition Node would fail to parse this.

The Condition Agent, however, intercepts the transcript seamlessly in the Realtime stream, calculates the semantic meaning using GPT/Claude, and routes the audio execution graph downwards into the appropriate path without requiring a hard-coded mapping logic.

[!CAUTION] Utilizing a Condition Agent introduces a slight LLM inference latency penalty on the call. Use determinist Condition Node paths whenever exact matching is possible (e.g. keypress responses) to maintain strict sub-400ms TTS latency.