How LLM streaming works on the wire
Comes up in: Frontend system design, AI system design, Web fundamentals
This concept is being written. It gets the full five steps — explanation, predict, interview angle, practice with tests, and review cards. Here is exactly what it will cover, so you can study it elsewhere in the meantime.
What it covers
- server-sent events format
- streaming event types in LLM APIs (message start, content block deltas, tool-use input deltas, stop reasons — using the Claude Messages API as the worked example)
- why a backend proxy is needed (never expose API keys)
- proxying streams through a Worker/Next route
Words you will know
What you will practise
- build a streaming proxy endpoint (mock provider in CI)
Interview questions this answers
- How would you stream an LLM response into a React UI?
- Why should the browser not call the LLM provider directly?
- How do you handle a stream that fails halfway?