add KV-cache and move system promt to the config

config.yml

@@ -1,6 +1,13 @@
 # LLM backend: "local" or "lmstudio"
 llm:
   backend: local        # change to "lmstudio" to use LM Studio instead
+  max_cache_tokens: 4096  # max KV-cache size per session (tokens); 0 to disable caching
+
+  system_prompt: >-
+    You are a helpful voice assistant. Keep your responses concise and natural
+    for spoken conversation. Respond in 1-3 short sentences.
+    Do not use markdown, bullet points, code blocks, emojis, or any
+    formatting that doesn't work in speech.
 
   # Settings used only when backend = "lmstudio"
   lmstudio: