Add ux profiles and advanced config block

src/aiprocess.py

@@ -69,6 +69,7 @@ class LlamaProcessor:
         lang: str = "en",
         *,
         dictionary_context: str = "",
+        profile: str = "default",
     ) -> str:
         request_payload: dict[str, Any] = {
             "language": lang,
@@ -87,6 +88,7 @@ class LlamaProcessor:
         }
         if _supports_response_format(self.client.create_chat_completion):
             kwargs["response_format"] = {"type": "json_object"}
+        kwargs.update(_profile_generation_kwargs(self.client.create_chat_completion, profile))
 
         response = self.client.create_chat_completion(**kwargs)
         return _extract_cleaned_text(response)
@@ -205,11 +207,25 @@ def _extract_cleaned_text(payload: Any) -> str:
 
 
 def _supports_response_format(chat_completion: Callable[..., Any]) -> bool:
+    return _supports_parameter(chat_completion, "response_format")
+
+
+def _supports_parameter(callable_obj: Callable[..., Any], parameter: str) -> bool:
     try:
-        signature = inspect.signature(chat_completion)
+        signature = inspect.signature(callable_obj)
     except (TypeError, ValueError):
         return False
-    return "response_format" in signature.parameters
+    return parameter in signature.parameters
+
+
+def _profile_generation_kwargs(chat_completion: Callable[..., Any], profile: str) -> dict[str, Any]:
+    normalized = (profile or "default").strip().lower()
+    if normalized != "fast":
+        return {}
+    if not _supports_parameter(chat_completion, "max_tokens"):
+        return {}
+    # Faster profile trades completion depth for lower latency.
+    return {"max_tokens": 192}
 
 
 def _llama_log_callback_factory(verbose: bool) -> Callable:

src/aman.py

@@ -233,6 +233,7 @@ class Daemon:
                     text,
                     lang=STT_LANGUAGE,
                     dictionary_context=self.vocabulary.build_ai_dictionary_context(),
+                    profile=self.cfg.ux.profile,
                 )
                 if ai_text and ai_text.strip():
                     text = ai_text.strip()