From 486ee37630c1c7ed1ae98373c0303fbde8c0c863 Mon Sep 17 00:00:00 2001
From: kobeng <38723692@qq.com>
Date: Sun, 12 Apr 2026 18:05:52 +0800
Subject: [PATCH] feat(llm): add stream mode support for proxies that only
 return content via streaming

Some OpenAI-compatible proxies (e.g. jarodfund.xyz) return empty content
in non-streaming mode. When stream=True is passed via OPENSPACE_LLM_CONFIG,
use litellm's stream_chunk_builder to collect streaming chunks into a
standard response.
---
 openspace/llm/client.py | 23 ++++++++++++++++++-----
 1 file changed, 18 insertions(+), 5 deletions(-)

diff --git a/openspace/llm/client.py b/openspace/llm/client.py
index 611cf87..867d28f 100644
--- a/openspace/llm/client.py
+++ b/openspace/llm/client.py
@@ -548,14 +548,27 @@ class LLMClient:
         - Total max time: timeout * max_retries + sum(retry_delays)
         """
         last_exception = None
+        use_stream = completion_kwargs.pop("stream", False)
+        if use_stream:
+            completion_kwargs["stream_options"] = {"include_usage": True}
         
         for attempt in range(self.max_retries):
             try:
-                # Add timeout to the completion call
-                response = await asyncio.wait_for(
-                    litellm.acompletion(**completion_kwargs),
-                    timeout=self.timeout
-                )
+                if use_stream:
+                    chunks = []
+                    async for chunk in await litellm.acompletion(stream=True, **completion_kwargs):
+                        chunks.append(chunk)
+                    if not chunks:
+                        raise ValueError("Stream returned no chunks")
+                    response = await asyncio.wait_for(
+                        asyncio.to_thread(litellm.stream_chunk_builder, chunks, completion_kwargs.get("messages", [])),
+                        timeout=self.timeout,
+                    )
+                else:
+                    response = await asyncio.wait_for(
+                        litellm.acompletion(**completion_kwargs),
+                        timeout=self.timeout
+                    )
                 return response
             except asyncio.TimeoutError:
                 self._logger.error(