Stream Blog

The future of software is conversational and interactive. For developers, unlocking this frontier means moving beyond traditional text inputs to agents that can seamlessly see, hear, and speak. Our goal is to demonstrate a powerful, flexible architecture that achieves this. This allows us to build truly expressive, realtime-latency AI applications. To illustrate, consider our core

Furnished Finder is the leading two-sided marketplace for monthly furnished rentals, connecting landlords with traveling professionals, remote workers, and relocating families seeking stays of 30 days or more. With a growing network of over 300,000 listings and more than 240,000 landlords nationwide, Furnished Finder connects over 6 million renters annually to real homes in real

Google dropped Gemini 3 Flash, a fast multimodal model that excels at video understanding, live frame analysis, and object detection. Plus, it’s cost-effective and offers low latency. In this quick demo, we use it to build a vision AI app in under five minutes that watches your camera feed in real time, accurately describes what

Two months after v0.2, we’re excited to share Vision Agents v0.3—our next significant milestone towards running agents in production at scale. While v0.2 introduced the foundation for building realtime multimodal AI agents, v0.3 takes these agents from prototype to production. This release brings the infrastructure you need to deploy agents at scale: HTTP APIs, observability,

Peerspace is the leading marketplace for booking unique spaces for meetings, productions, and events. The platform connects guests with hosts through real-time, in-app messaging, enabling seamless coordination, faster bookings, and stronger trust on both sides of the marketplace. For Peerspace, keeping conversations inside the platform is a strategic priority. In-app messaging reduces reliance on third-party

xAI’s Grok-4 delivers strong reasoning with a 256k context window, native tool use, and multimodal support. We love it for natural, low-latency voice conversations. Pair it with Fish Audio’s high-quality, expressive TTS (known for realistic prosody, emotion control, and voice cloning via short references) and Deepgram’s fast, accurate STT, and you get a custom voice

Every vision-language model tutorial shows you the same thing: send an image to GPT-4o, get a description back. Ten lines of Python. Done. response = client.chat.completions.create(     model="gpt-4o",     messages=[{         "role": "user",         "content": [             {"type": "image_url", "image_url": {"url": f"data:image/jpeg;base64,{img_b64}"}},             {"type": "text", "text": "What's in this image?"}         ]     }] ) Real applications need something different. A security camera

Over the last few years, developers have gone from using language models for text-only chat to relying on them as general-purpose perception systems. You’re not only building chatbots; you’re building apps that use text, audio, and vision to understand and act on the world around them. GPT-4o is the most capable step yet: a single