streaming

The streaming skill provides guidance for implementing and troubleshooting streaming communication between AI backends and assistant-ui using the assistant-stream package. It is designed for developers building custom streaming endpoints rather than using the Vercel AI SDK workflow. The skill focuses on creating streaming responses, handling streamed content parts, selecting appropriate wire protocols, and ensuring that assistant responses are delivered incrementally to the user interface. It also helps diagnose common streaming issues such as missing updates, malformed server-sent events, and rendering problems with streamed content.

Core capabilities include creating custom streaming responses, emitting streamed content such as text, reasoning, sources, files, and tool-call data, and attaching tool results to active streams. The skill covers both the AI SDK Data Stream format and the native Assistant Transport format, along with the encoders and decoders used to serialize and deserialize stream data. It also supports stream inspection and debugging through stream event processing, including events such as part-start, text-delta, result, step lifecycle events, message completion, and errors. Additionally, it explains how streamed chunks can be integrated into runtime systems such as useLocalRuntime.

Typical use cases include building custom AI backend integrations, implementing real-time chat experiences, debugging server-sent event streams, validating wire protocol behavior, and ensuring tool calls render correctly in assistant-ui applications. The primary audience includes JavaScript and TypeScript developers, frontend engineers working with React-based assistant-ui integrations, and backend developers responsible for streaming AI responses from custom services.