Skip to content

l11-api-streaming.en.md

Latest commit

 

History

History
48 lines (31 loc) · 1.48 KB

l11-api-streaming.en.md

File metadata and controls

48 lines (31 loc) · 1.48 KB

Layer 11: API Streaming And The Event Model

中文 | English

Core Question

How does Claude Code transform the raw API stream into a unified runtime event model?

Review First

  • l2-agent-loop
  • l8-streaming

Read First

  • services/api/claude.ts
  • query.ts

Search Anchors

  • queryModelWithStreaming
  • api_request_sent
  • first_chunk
  • stream_request_start
  • message_start

What To Notice

  • how request-sent, headers-received, and first-chunk timings are measured
  • why query.ts emits stream_request_start before the stream is consumed
  • why watchdogs, stall detection, and fallback timeouts live close to the API layer

Source Evidence

  • services/api/claude.ts:752: export async function* queryModelWithStreaming
  • services/api/claude.ts:1805-1807: api_request_sent
  • services/api/claude.ts:1971-1973: first_chunk
  • query.ts:337: stream_request_start

What You Should Actually Learn

Claude Code streaming is not just text chunking. It is a runtime model that unifies network state, event boundaries, recovery behavior, and latency measurement into one consumable stream. The lesson here is not “how to use the Anthropic SDK,” but how to define event semantics cleanly.

Questions To Answer

  1. Why does services/api/claude.ts care about stalls and idle timeouts?
  2. What is the most important contract between query.ts and the API layer?
  3. If the underlying stream dies, why is a generic exception not enough?