Streaming an LLM reply to the browser with fetch: read response.body with a TextDecoder
EventSource only does GET, so for a POST with a prompt use fetch and read the stream: const reader = res.body.getReader(); const dec = new TextDecoder(); while (true) { const { done, value } = await reader.read(); if (done) break; append(dec.decode(value, { stream: true })); }. Chunks can split a line or a multi-byte character, which is why stream: true matters; if the server sends SSE-formatted lines, buffer until a blank line before parsing. Proxies that buffer responses make it arrive all at once.
Two small additions: check res.ok before reading the body (an error response streams fine and you end up appending an error page), and call dec.decode() once with no arguments after the loop to flush any trailing bytes of a multi-byte character held back by stream: true.