Tutorials

A camera used to be proof enough that the person on the other end was who they said they were. This year, that assumption ran out. This piece walks through the four security layers a video call actually has, transport encryption, media encryption, account authorization, and human verification, and what closing the last gap looks like in practice.

A few years ago, building an AI voice agent took an entire team and a full codebase. Now, if you're a Python developer, it's a single-file problem. This tutorial builds a real-time voice agent with Python and WebRTC that you can talk to in your browser, and explains why transport, not the AI, is the hard part.

Building a "Zoom inside your app" with classroom rules from raw WebRTC means months of work on signaling, SFUs, device handling, and reconnection logic before you write a single classroom feature. This tutorial builds it in an afternoon with Stream's React Video SDK, a virtual classroom where a teacher runs the class from a laptop and students join from their phones, with hand-raising, a moderated microphone, and a live roster.

Build a voice shopping concierge for a multi-seller marketplace using Kimi K2.5, Deepgram STT, and Vision Agents. Runnable Python, function calling, Stream WebRTC.

Discover how to integrate and use local, free, and open-source Core AI models with Stream Chat.

The design process behind one-shot generating VinylStream, a live shopping app for buying and selling vinyl - and why a good design document, not the fanciest model, is what makes AI app generation actually work.

Build on-device suggested replies in a SwiftUI chat app with Apple's Foundation Models and Stream Chat. Generation runs entirely on the device: no token cost, no network round-trip.

Build a Python voice agent with Vision Agents that places an outbound call to a cardholder, reads back a suspicious transaction, and either clears it or freezes the card - using Deepgram, an LLM, Cartesia, and Twilio.