streaming-api-patterns_skill

Overview

This skill teaches robust real-time streaming patterns using Server-Sent Events (SSE), WebSockets, and the ReadableStream API in TypeScript. It focuses on practical techniques for LLM streaming, backpressure control, reconnection strategies, and production-ready performance tuning. You will get clear patterns for one-way, bidirectional, and high-throughput streaming suited for 2025+ real-time applications.

How this skill works

Implementations show how to expose ReadableStream-based endpoints that push SSE frames, run WebSocket servers for two-way messaging, and produce controlled streams that respect consumer backpressure. The skill demonstrates reading streamed LLM responses, encoding tokens into SSE frames, and client-side readers that decode and process incremental data. It also covers reconnection logic, keepalive heartbeats, and flow control to avoid buffer exhaustion and ensure stable delivery.

When to use it

• Deliver incremental LLM responses to a web UI (token-by-token streaming)
• Broadcast real-time notifications, analytics, or market data from server to many clients
• Support chat or collaborative editors with low-latency bidirectional updates
• Stream progress for long-running tasks and provide live progress bars
• Process large datasets with backpressure-aware pipelines on the server

Best practices

• Use SSE for simple server-to-client streams and WebSockets for bidirectional needs
• Implement exponential backoff reconnection and send periodic keepalive frames
• Respect backpressure via ReadableStream controllers and pause production when consumers are slow
• Compress and batch small messages; prefer binary formats (MessagePack/Protobuf) for heavy payloads
• Validate and sanitize all incoming WebSocket messages and monitor connection counts

Example use cases

• Streaming chat UI that renders tokens as the LLM returns them
• Live dashboard feeding fast telemetry (stock ticks, metrics) to many viewers
• Real-time collaborative text editor syncing changes via WebSockets
• Backend pipeline that streams large file transforms with controlled flow
• Progress streaming for background jobs (builds, data processing) to client UIs