Pythia provider
Pythia is a self-hosted Realtime provider (pythia_realtime) maintained by KI Kombinat for data-residency or cost-sensitive deployments where public cloud STT/TTS is not acceptable.
Fields
| Field | Required | Description |
|---|---|---|
| URL | Yes | WebSocket URL for your Pythia Realtime server |
| API key | No | Optional if your deployment requires auth |
| Model, voice, transcription, VAD | No | Same advanced tuning pattern as OpenAI Realtime |
Example:
wss://pythia-echo.example.com/v1/realtime
wss://10.0.5.12:9090/v1/realtime
Pythia Realtime defaults to G.711-oriented audio (alaw / telephony wire formats) in the provider AI codec profile — align endpoint and carrier codecs when bridging PSTN.
When to use Pythia
| Scenario | Notes |
|---|---|
| On-prem unified voice | Single URL replaces separate STT + LLM + TTS vendors |
| Lab / echo testing | Point at a internal mock that speaks the Pythia Realtime protocol |
| Private IP only | Set URL to wss://<ip>:<port>/…; configure Voice routing and NO_PROXY as needed |
See also
- Custom STT/TTS endpoints
- Providers overview
- Azure — alternative modular STT/TTS with per-stage endpoints