Vercel
WebSockets on Vercel: native support, limits, and your options
Vercel Functions now support WebSockets natively, with real limits. See your options for realtime and AI apps, and when native support isn't enough.
Vercel WebSockets vs Ably: fan-out, presence, and ordering compared
Vercel Functions can now handle WebSocket connections. Here's how fan-out, presence, and message ordering compare to Ably, and where each one stops.
Durable sessions for Vercel AI SDK applications
Vercel AI SDK's SSE transport breaks in production: proxy buffering, no reconnect, serverless limits. ChatTransport makes it swappable. Options compared.
Why Vercel AI SDK can't stream to multiple devices
SSE is a one-to-one connection. Delivering the same AI generation to multiple clients or devices requires a broadcast layer that SSE cannot provide.
Vercel AI SDK resumable-stream: what it covers and what it doesn't
Vercel's resumable-stream covers page reloads only. Tab switches, mobile backgrounding, and device switches lose the stream. Also incompatible with stop().
Vercel AI SDK ChatTransport: implementing a custom WebSocket transport
ChatTransport in Vercel AI SDK 5 lets you replace the default HTTP transport with WebSockets. Application code, agents, and UI stay unchanged.
Why Vercel AI SDK stop() doesn't cancel the stream
stop() in Vercel AI SDK closes the connection, not the generation. The abort signal is a separate HTTP request and may never reach the generating process.
WebSockets on Vercel: why serverless functions can't host them
Vercel serverless functions can't host WebSocket connections, even with Fluid Compute. Options and how to connect a WebSocket provider to Vercel AI SDK.
Why Vercel AI SDK stream errors fail silently in production
Why Vercel AI SDK stream errors fail silently in production, three SDK patterns to catch them, and what a durable transport layer adds.
Why AI chat history disappears between sessions
Vercel AI SDK stores messages in component state. When the page reloads or the user returns later, the history is gone. How to add durable persistence to AI applications.