Filter litellm models to chat mode only

- Add _litellm_chat_prices_cache for filtered models
- Rename get_litellm_prices() to get_litellm_prices_raw()
- Create wrapper get_litellm_prices() that filters mode='chat'
- All model lookups now return only chat models

app.py

@@ -50,6 +50,7 @@ def _log_unhandled(exc_type, exc_value, exc_traceback):
 sys.excepthook = _log_unhandled
 
 _litellm_prices_cache = None
+_litellm_chat_prices_cache = None
 _trajectories_cache = {}
 _calculated_tokens_cache = {}
 _trajectory_steps_cache = {}
@@ -391,13 +392,8 @@ def load_all_trajectory_steps(folder: str) -> dict[str, list[dict]]:
     return result
 
 
-def get_litellm_model_list() -> list[str]:
-    """Get list of model names from litellm prices"""
-    prices = get_litellm_prices()
-    return sorted(prices.keys())
-
-
-def get_litellm_prices() -> dict:
+def get_litellm_prices_raw() -> dict:
+    """Get raw litellm prices (all modes, unfiltered)"""
     global _litellm_prices_cache
     if _litellm_prices_cache is not None:
         return _litellm_prices_cache
@@ -421,6 +417,26 @@ def get_litellm_prices() -> dict:
     return _litellm_prices_cache
 
 
+def get_litellm_prices() -> dict:
+    """Get litellm prices filtered to chat models only"""
+    global _litellm_chat_prices_cache
+    if _litellm_chat_prices_cache is not None:
+        return _litellm_chat_prices_cache
+
+    raw_prices = get_litellm_prices_raw()
+    _litellm_chat_prices_cache = {
+        k: v for k, v in raw_prices.items()
+        if isinstance(v, dict) and v.get("mode") == "chat"
+    }
+    return _litellm_chat_prices_cache
+
+
+def get_litellm_model_list() -> list[str]:
+    """Get list of chat model names from litellm prices"""
+    prices = get_litellm_prices()
+    return sorted(prices.keys())
+
+
 def normalize_model_name(name: str) -> str:
     """Normalize model name for comparison: lowercase, remove separators"""
     return re.sub(r'[-_./]', '', name.lower())