perf: Haiku for simple queries, Sonnet for complex — portfolio/property/market queries now ~2s, strategy/life-decision ~10s

Made-with: Cursor

agent/graph.py

@@ -76,6 +76,33 @@ try:
 except ImportError:
     _RE_STRATEGY_AVAILABLE = False
 
+# Model selection constants
+FAST_MODEL = "claude-haiku-4-5-20251001"
+SMART_MODEL = "claude-sonnet-4-20250514"
+
+# Query types that need Sonnet for quality
+COMPLEX_QUERY_TYPES = {
+    "life_decision",
+    "family_planner",
+    "wealth_gap",
+    "wealth_down_payment",
+    "wealth_job_offer",
+    "wealth_global_city",
+    "wealth_portfolio_summary",
+    "equity_unlock",
+    "real_estate_detail",
+    "real_estate_snapshot",
+    "real_estate_search",
+    "real_estate_compare",
+}
+
+def get_model_for_query(query_type: str) -> str:
+    """Returns appropriate model based on query complexity."""
+    if query_type in COMPLEX_QUERY_TYPES:
+        return SMART_MODEL
+    return FAST_MODEL
+
+
 SYSTEM_PROMPT = """You are a portfolio analysis assistant integrated with Ghostfolio wealth management software.
 
 REASONING PROTOCOL — silently reason through these four steps BEFORE writing your response.
@@ -2393,8 +2420,10 @@ async def format_node(state: AgentState) -> AgentState:
                 ),
             })
             try:
+                _qt = state.get("query_type", "portfolio")
+                _model = get_model_for_query(_qt)
                 response_obj = client.messages.create(
-                    model="claude-sonnet-4-20250514",
+                    model=_model,
                     max_tokens=800,
                     system=SYSTEM_PROMPT,
                     messages=api_messages_ctx,
@@ -2531,8 +2560,10 @@ async def format_node(state: AgentState) -> AgentState:
     actual_input_tokens: int | None = None
     actual_output_tokens: int | None = None
     try:
+        _qt = state.get("query_type", "portfolio")
+        _model = get_model_for_query(_qt)
         response_obj = client.messages.create(
-            model="claude-sonnet-4-20250514",
+            model=_model,
             max_tokens=800,
             system=SYSTEM_PROMPT,
             messages=api_messages,