feat: add AI_VOICE_AI_TIMEOUT_SECONDS for configurable voice timeout

services/ai_orchestrator_service/app.py

@@ -1475,6 +1475,14 @@ def _ai_timeout_seconds() -> float:
     return max(3.0, _float_env("AI_TIMEOUT_SECONDS", 20.0))
 
 
+def _ai_voice_timeout_seconds() -> float:
+    return max(3.0, _float_env("AI_VOICE_AI_TIMEOUT_SECONDS", _ai_timeout_seconds()))
+
+
+def _ai_decision_max_tokens() -> int:
+    return max(64, int(_float_env("AI_DECISION_MAX_TOKENS", 500)))
+
+
 def _ai_telegram_enabled() -> bool:
     return _bool_env("AI_TELEGRAM_ENABLED", False)
 
@@ -2339,17 +2347,23 @@ def _openai_prompt(
     ]
 
 
-def _request_structured_model_decision(messages: list[dict[str, str]]) -> dict[str, Any]:
+def _request_structured_model_decision(
+    messages: list[dict[str, str]],
+    *,
+    timeout_seconds: float | None = None,
+) -> dict[str, Any]:
     if not _ai_api_base() or not _ai_api_key():
         raise RuntimeError("AI_API_BASE / AI_API_KEY are required for openai_compatible provider")
     started = time.perf_counter()
     payload = {
         "model": _ai_model(),
         "temperature": 0.2,
+        "max_tokens": _ai_decision_max_tokens(),
         "response_format": {"type": "json_object"},
         "messages": messages,
     }
-    with httpx.Client(timeout=_ai_timeout_seconds()) as client:
+    effective_timeout = timeout_seconds if timeout_seconds is not None else _ai_timeout_seconds()
+    with httpx.Client(timeout=effective_timeout) as client:
         response = client.post(
             f"{_ai_api_base()}/chat/completions",
             json=payload,

services/ai_orchestrator_service/voice.py

@@ -1664,7 +1664,8 @@ def _voice_llm_decision(
                 name_value=name_value,
                 name_status=name_status,
                 operator_config=operator_config,
-            )
+            ),
+            timeout_seconds=app._ai_voice_timeout_seconds(),
         )
     except Exception:
         return None