Stream (don't buffer) the chat response so members see tokens immediately

app/main.py

@@ -26,7 +26,7 @@ from email.mime.multipart import MIMEMultipart
 import httpx
 import jwt
 from fastapi import FastAPI, Request, Response, HTTPException
-from fastapi.responses import JSONResponse
+from fastapi.responses import JSONResponse, StreamingResponse
 
 logging.basicConfig(level=logging.INFO)
 logger = logging.getLogger("member-portal")
@@ -921,29 +921,46 @@ async def inject_key(request: Request, path: str):
     if not member_key:
         raise HTTPException(403, "No active membership key")
 
-    # Forward the request to LiteLLM with the member's key.
+    # Forward the request to LiteLLM with the member's key, STREAMING the
-    # Use a generous timeout: model generations can take well over the httpx
+    # response back token-by-token (or byte-by-byte) rather than buffering it.
-    # default of 5s, and the previous code buffered the full response which
+    # Buffering collapsed the model's stream and handed OpenWebUI the full
-    # caused httpx.ReadTimeout on any non-trivial generation.
+    # answer in one blob, so members saw nothing until generation finished —
+    # which made chat feel slow even though time-to-first-byte was fine.
     body = await request.body()
     headers = dict(request.headers)
     headers["authorization"] = f"Bearer {member_key}"
     headers.pop("host", None)
     headers.pop("content-length", None)
+    headers.pop("accept-encoding", None)  # let httpx handle decompression
 
     timeout = httpx.Timeout(connect=15.0, read=600.0, write=30.0, pool=15.0)
-    async with httpx.AsyncClient(timeout=timeout) as client:
-        upstream = await client.request(
-            method=request.method,
-            url=f"{LITELLM_BASE}/v1/{path}",
-            headers=headers,
-            content=body,
-        )
 
-    return Response(
+    # Send the request upstream, but don't buffer the body — use a stream.
-        content=upstream.content,
+    client = httpx.AsyncClient(timeout=timeout)
+    upstream_req = client.build_request(
+        method=request.method,
+        url=f"{LITELLM_BASE}/v1/{path}",
+        headers=headers,
+        content=body,
+    )
+    upstream = await client.send(upstream_req, stream=True)
+
+    async def gen():
+        try:
+            async for chunk in upstream.aiter_raw():
+                yield chunk
+        finally:
+            await upstream.aclose()
+            await client.aclose()
+
+    # Pass through the content-type (critical for SSE streaming) and status.
+    return StreamingResponse(
+        gen(),
         status_code=upstream.status_code,
-        headers={"content-type": upstream.headers.get("content-type", "application/json")},
+        headers={
+            "content-type": upstream.headers.get("content-type", "application/json"),
+            "cache-control": "no-cache",
+        },
     )