Streaming/SSE fails silently in GUI

[stgreenb]

Goose Desktop fails to display streaming responses from custom OpenAI‑compatible provider, even though the provider returns valid SSE and non‑streaming works correctly.

This my be related to #3786 and #4685 (custom provider integration issues), but specific to streaming mode.

Environment

• OS: Windows Desktop app
• Goose: Latest Desktop build (as of 2025‑12‑18)
• Mode: Autonomous

Provider: Custom OpenAI‑compatible /v1/chat/completions endpoint

Steps to reproduce

1. Configure custom provider with engine: "openai", supports_streaming: true
2. Send prompt in Autonomous mode
3. Observe: Request sent with stream: true, provider returns valid SSE chunks, Goose UI shows nothing
4. Change supports_streaming: false, resend prompt
5. Observe: UI displays full response correctly

What works

• Non‑streaming responses from same provider
• SSE stream from provider (verified via Bruno HTTP client)
• Goose receives the request (confirmed in logs)

What doesn't work

• Streaming responses displayed in Goose UI
• Silent failure—no error message or timeout
• Example valid SSE response from provider

Example streamed response (sanitized) This is an example of what the provider returns when stream: true is set. Each line is a separate SSE data: frame:

data:{"id":"chat","object":"chat.completion.chunk","created":1766073356,"choices":[{"index":0,"delta":{"role":"assistant","content":""}}],"model":"glm-4.6"}
data:{"id":"chat","object":"chat.completion.chunk","created":1766073356,"choices":[{"index":0,"delta":{"content":"\nHello","reasoning_content":""}}],"model":"glm-4.6"}
...
data:{"id":"chat","object":"chat.completion.chunk","created":1766073357,"choices":[{"index":0,"delta":{"content":" to help you with today?","reasoning_content":""}}],"model":"glm-4.6"}

data:{"id":"chat","object":"chat.completion.chunk","created":1766073357,"usage":{"prompt_tokens":7840,"completion_tokens":49,"total_tokens":7889},"choices":[{"index":0,"finish_reason":"stop","delta":{"content":""}}],"model":"glm-4.6"}

Note:

Every chunk has object: "chat.completion.chunk" and choices[0].delta.content, matching the usual OpenAI streaming shape. The last chunk includes finish_reason: "stop" and usage. When replayed with another HTTP client, the stream is readable and decodes to the expected text.