AI providers
Providers hold vendor-specific model, voice, language, transcription, and audio settings. Credentials are stored in Vault and referenced by provider objects.
The current control plane supports OpenAI, Deepgram, and xAI provider types. Runtime support is divided by modality:
| Provider | Realtime | STT | LLM | TTS |
|---|---|---|---|---|
| OpenAI | Yes | — | Yes | — |
| Deepgram | Yes | Yes | — | Yes |
| xAI | Yes | — | — | — |
A transport supplies the network endpoint and timeout behavior. A modality binds one provider to one transport for a specific role. This separation allows several provider accounts, models, or connection paths to coexist in the same tenant.
Provider objects can be reused across Bots, but each dependency must be assigned to every Bot that consumes it.