Fix injector ReadTimeout (600s read timeout); switch default model to deepseek-v4-1-flash

app/main.py

@@ -79,7 +79,7 @@ OC_PERSONAL_TOKEN = os.environ.get("OC_PERSONAL_TOKEN", "")
 MEMBER_BUDGET = float(os.environ.get("MEMBER_BUDGET", "15.0"))
 
 # Models every member key/team may access.
-MEMBER_MODELS = ["deepseek-v4-flash", "gpt-oss-120b", "glm-5-3-flash"]
+MEMBER_MODELS = ["deepseek-v4-1-flash", "gpt-oss-120b", "glm-5-3-flash"]
 
 # The "members" group in Cloudron (group-based access control).
 # Members are assigned to this group, which grants access to the chat app.
@@ -921,14 +921,18 @@ 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.
+    # Use a generous timeout: model generations can take well over the httpx
+    # default of 5s, and the previous code buffered the full response which
+    # caused httpx.ReadTimeout on any non-trivial generation.
     body = await request.body()
     headers = dict(request.headers)
     headers["authorization"] = f"Bearer {member_key}"
     headers.pop("host", None)
     headers.pop("content-length", None)
 
-    async with httpx.AsyncClient() as client:
+    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}",