# Streaming (SSE) For text and JSON skills, HUSK can stream the kernel's output as it is produced instead of buffering the whole result. This is ideal for long-running kernels and for token-by-token output from a model-backed script. ## Asking for a stream Add `?stream=1` to the URL, or send `Accept: text/event-stream`: ```bash curl -N -X POST 'http://localhost:3000/skills/uppercase?stream=1' --data 'hello' ``` ``` event: stdout data: HELLO event: done data: {"ok":true,"exitCode":0} ``` The response is a [Server-Sent Events](https://developer.mozilla.org/docs/Web/API/Server-sent_events) stream. ## Event types | Event | `data` | | -------- | ---------------------------------------------- | | `stdout` | A chunk of the kernel's standard output. | | `stderr` | A chunk of the kernel's standard error. | | `done` | `{ ok, exitCode }` once the kernel exits. | | `error` | A message if the request could not be handled. | Chunks arrive as the kernel writes them, so a kernel that prints incrementally streams incrementally. ## Consuming a stream In the browser: ```js const res = await fetch('/skills/summarize?stream=1', { method: 'POST', body: text }); const reader = res.body.getReader(); const decoder = new TextDecoder(); for (;;) { const { value, done } = await reader.read(); if (done) break; console.log(decoder.decode(value)); // raw SSE frames } ``` Any SSE client (`EventSource`-style parsing) works the same way. :::note Streaming applies to `text` and `json` skills. `file` skills always return their file as a single response, so a stream request on a file skill falls back to the normal buffered response. ::: ## Writing a streaming kernel Nothing special is required - just write to stdout incrementally and flush. For example, a Python kernel: ```python #!/usr/bin/env python3 import sys, time for word in sys.stdin.read().split(): print(word.upper(), flush=True) time.sleep(0.1) ``` Each `print(..., flush=True)` becomes an `event: stdout` frame as it happens.