Skip to main content
Stream’s Python AI SDK ships with a growing catalogue of plugins that connect third-party AI services to your live video calls. Each plugin wraps a specific AI provider and exposes a unified API so you can swap vendors without rewriting business logic.

​
Why Use a Plugin

  • Quality & speed: Leverage specialised models (e.g. Deepgram’s high-accuracy STT, ElevenLabs’ realistic TTS) without hosting anything yourself.
  • Drop-in architecture: Plugins of the same type inherit from the same base class with a consistent interface (e.g. STT plugins implement process_audio() and close(), emit transcript events). You can swap implementations in minutes.
  • Runs inside your call: The SDK streams PCM frames directly to the provider in real-time, then emits SDK events that can be listened for and acted upon.

​
What can you build?

CapabilityExample providers
Speech-to-Text (STT)Deepgram, Moonshine (local)
Text-to-Speech (TTS)ElevenLabs, Kokoro
Voice Activity Detection (VAD)Silero
Combine them to create richer pipelines; e.g. VAD → STT → Moderation → TTS for a real-time, policy-aware voice agent. See the other pages in this section for our individual third-party integrations. We’ll add more over time, and you can even write your own!

​
Creating Your Own Plugins

You can absolutely write your own plugins to connect other AI providers to the AI Python SDK! Follow this guide to learn how.
⌘I