fix: make AI parser async to prevent Discord heartbeat blocking

ai/parser.py

@@ -8,6 +8,7 @@ Config-driven via ai_config.json. Supports:
 - Conversation context for multi-turn interactions
 """
 
+import asyncio
 import json
 import os
 import re
@@ -37,8 +38,8 @@ def _extract_json_from_text(text):
     return None
 
 
-def _call_llm(system_prompt, user_prompt):
-    """Call OpenAI-compatible API and return the response text."""
+def _call_llm_sync(system_prompt, user_prompt):
+    """Synchronous LLM call - runs in thread pool to avoid blocking."""
     try:
         response = client.chat.completions.create(
             model=AI_CONFIG["model"],
@@ -64,7 +65,12 @@ def _call_llm(system_prompt, user_prompt):
         return None
 
 
-def parse(user_input, interaction_type, retry_count=0, errors=None, history=None):
+async def _call_llm(system_prompt, user_prompt):
+    """Async wrapper for LLM call - prevents Discord heartbeat blocking."""
+    return await asyncio.to_thread(_call_llm_sync, system_prompt, user_prompt)
+
+
+async def parse(user_input, interaction_type, retry_count=0, errors=None, history=None):
     """
     Parse user input into structured JSON using LLM.
 
@@ -111,14 +117,14 @@ def parse(user_input, interaction_type, retry_count=0, errors=None, history=None
             f"\n\nPrevious attempt had errors: {errors}\nPlease fix and try again."
         )
 
-    response_text = _call_llm(prompt_config["system"], user_prompt)
+    response_text = await _call_llm(prompt_config["system"], user_prompt)
     if not response_text:
         return {"error": "AI service unavailable", "user_input": user_input}
 
     try:
         parsed = json.loads(response_text)
     except json.JSONDecodeError:
-        return parse(
+        return await parse(
             user_input,
             interaction_type,
             retry_count + 1,