Product
Voice pipelines
Speed, Smart, Ultra, and Custom stacks on the Pipecat telephony pipeline.
Modes
Speed — lowest turn latency. Smart — stronger reasoning with a modest latency trade. Ultra — speech-to-speech with mid-call tool/KB guidance (may require paid plan).Custom multi-STT
Pick AssemblyAI, Gladia, Deepgram, or Whisper variants in the Transcriber panel. Pair with your preferred LLM and TTS.
Runtime hardening
Live calls run on Pipecat with Deepgram STT, fractional LLM→TTS streaming, and PSTN denoise. The platform audits custom frame processors at boot so a swallowed StartFrame or TTSSpeakFrame cannot ship silently. After the greeting, turn-start interruptions are disabled on telephony so user speech reaches STT finals before the LLM answers. Ambient beds (10 selectable scenes — office, cafe, call center, rain, …) mix under TTS when chosen in Call Settings → Background ambience.