Skip to main content
Version: 0.9.16 — archived documentation

OpenAI provider

OpenAI providers cover Realtime (unified voice), Chat (pipeline LLM), and TTS (pipeline speech synthesis). OpenAI does not appear as a modular STT stage — use Azure or Inworld STT in pipeline mode, or OpenAI Realtime for unified speech.

Modes​

Flow modeProvider typesPrimary use
Realtimeopenai_realtimeSingle connection for STT + LLM + TTS
Pipelineopenai_chat, openai_ttsText LLM and/or speech output with separate STT upstream

Realtime (openai_realtime)​

FieldRequiredDescription
ModelNogpt-realtime-* family (defaults to platform catalogue)
API keyYesTeam variable, e.g. {{team.openaiDefaultKey}}
VoiceNoOpenAI voice id (alloy, shimmer, …)
SpeedNoPlayback speed multiplier
Custom URLNoAdvanced → override WebSocket/API base (default: OpenAI production)
Transcription / VAD / noise reductionNoAdvanced → realtime input tuning

Transcription model and VAD settings apply to the realtime session's inbound audio, not to a separate STT provider.

Chat / LLM (openai_chat)​

FieldRequiredDescription
ModelNogpt-* chat models
API keyYesTeam OpenAI key variable
Base URLNoAdvanced → custom API host (default https://api.openai.com/v1)
Temperature, max tokens, penalties, tool choiceNoAdvanced LLM

TTS (openai_tts)​

FieldRequiredDescription
ModelNotts-* models
VoiceNoOpenAI TTS voice
API keyYesTeam OpenAI key variable
Speed, instructionsNoAdvanced TTS

The schema supports baseUrl for OpenAI TTS; if you need a non-default host, set it in the source view under providers.tts.baseUrl.

Pipeline TTS participates in the modular speech playback queue. Realtime audio is streamed by the vendor connection instead.

Custom endpoints​

See Custom STT/TTS endpoints — Realtime Custom URL and Chat Base URL accept hostnames and IPs for private gateways or compatible OpenAI-style APIs.

See also​