Plugins & Providers

Plugins & Providers

SIPHON is designed so you can choose the providers you want for:

  • LLM (reasoning / dialogue)
  • STT (speech-to-text)
  • TTS (text-to-speech)

This page explains the idea behind plugins and how to think about provider choices.

Why plugins exist

Voice systems are multi-provider by nature:

  • you might want one vendor for transcription and another for speech
  • you might swap LLM providers for cost/latency reasons
  • you might change voices without changing the rest of your application

Plugins let you swap components without changing your call routing logic.

The three core components

LLM

The LLM controls:

  • how the agent reasons
  • how it responds
  • when to use tools

STT

STT controls:

  • how quickly user speech becomes text
  • accuracy across accents/noisy calls
  • end-of-utterance behavior (affects turn taking)

TTS

TTS controls:

  • voice identity
  • latency
  • speaking rate/quality

Recommended approach

  • Start with a simple, known-good combination.
  • Then optimize based on your constraints:
    • latency
    • cost
    • accuracy
    • voice quality

Where to learn the practical API

Next