feat(avatar): add private vision chat support

digital-avatar-app/backend/services/chat_model_config.py

@@ -16,6 +16,10 @@ class ChatModelConfig:
     model: str
     max_tokens: int
     timeout_seconds: float
+    vision_model: str
+    ocr_model: str
+    vision_max_tokens: int
+    vision_timeout_seconds: float
     source: str
 
 
@@ -33,6 +37,10 @@ def _environment_config() -> ChatModelConfig:
         model=os.getenv("CHAT_MODEL", "qwen-plus"),
         max_tokens=max(128, int(os.getenv("CHAT_MAX_OUTPUT_TOKENS", "1024"))),
         timeout_seconds=max(5.0, float(os.getenv("CHAT_TIMEOUT_SECONDS", "30"))),
+        vision_model=os.getenv("VISION_MODEL", "qwen3.6-flash"),
+        ocr_model=os.getenv("VISION_OCR_MODEL", "qwen-vl-ocr"),
+        vision_max_tokens=max(256, int(os.getenv("VISION_MAX_OUTPUT_TOKENS", "2048"))),
+        vision_timeout_seconds=max(10.0, float(os.getenv("VISION_TIMEOUT_SECONDS", "90"))),
         source="environment",
     )
 
@@ -60,6 +68,20 @@ def _fetch_runtime_config() -> ChatModelConfig | None:
         model=model,
         max_tokens=max(128, int(payload.get("max_tokens") or 1024)),
         timeout_seconds=max(5.0, float(payload.get("timeout_seconds") or 30)),
+        vision_model=str(
+            payload.get("vision_model")
+            or os.getenv("VISION_MODEL", "qwen3.6-flash")
+        ),
+        ocr_model=str(
+            payload.get("ocr_model")
+            or os.getenv("VISION_OCR_MODEL", "qwen-vl-ocr")
+        ),
+        vision_max_tokens=max(
+            256, int(os.getenv("VISION_MAX_OUTPUT_TOKENS", "2048"))
+        ),
+        vision_timeout_seconds=max(
+            10.0, float(os.getenv("VISION_TIMEOUT_SECONDS", "90"))
+        ),
         source="admin",
     )