Streaming - Morse

Set "stream": true to receive the reply as the model writes it. Morse answers with Content-Type: text/event-stream. Each event is a line data: <json> with one chat.completion.chunk, and the stream ends with data: [DONE].

curl -N https://api.trymorse.ai/v1/chat/completions \
  -H "Authorization: Bearer $MORSE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.3",
    "messages": [{"role": "user", "content": "Write a haiku about rain."}],
    "stream": true,
    "stream_options": {"include_usage": true}
  }'

The -N flag stops curl from buffering the events. With the Python SDK:

stream = client.chat.completions.create(
    model="glm-5.3",
    messages=[{"role": "user", "content": "Write a haiku about rain."}],
    stream=True,
    stream_options={"include_usage": True},
)
for chunk in stream:
    if chunk.choices:
        print(chunk.choices[0].delta.content or "", end="")

Get the token counts

Morse sends a final chunk with usage only when the request sets stream_options.include_usage to true. No other chunk carries usage. The final chunk has an empty choices list.

Ignore the padding

Each chunk has a top-level obfuscation field, a random string of 16 to 64 characters that pads the chunk. It holds no data. Ignore it. On some reasoning models, Morse also sends a : keepalive comment every 25 seconds while the model thinks. Clients that follow the server-sent events specification ignore comments.

Handle an error mid-stream

If the model fails after the stream starts, the status is already 200. Morse then sends one event with the error envelope and ends the stream without data: [DONE]:

{"error":{"message":"The upstream model server failed.","type":"upstream_error","param":null,"code":"upstream_error"}}

Treat a stream that ends without data: [DONE] as incomplete.

​Get the token counts

​Ignore the padding

​Handle an error mid-stream

Get the token counts

Ignore the padding

Handle an error mid-stream