small refactor of the structure

gateway/dependencies.py

@@ -0,0 +1,36 @@
+from fastapi import Request
+from openai import OpenAI
+
+from src.graph import create_agent_graph
+
+
+def get_llm_client(request: Request) -> OpenAI:
+    """FastAPI dependency — returns the singleton OpenAI client from app.state."""
+    return request.app.state.llm_client
+
+
+def get_agent_graph(agent_id: str, request: Request):
+    """
+    FastAPI dependency — returns the compiled LangGraph graph for *agent_id*.
+
+    Graphs are lazily compiled on first use and cached on app.state so each
+    agent's graph is only built once per process lifetime.
+    """
+    cache: dict = request.app.state.agent_graphs
+
+    if agent_id not in cache:
+        from agents import get as get_agent
+
+        agent = get_agent(agent_id)
+        if agent is None:
+            # Fall back to the naked agent if the requested one doesn't exist
+            agent_id = "naked"
+            agent = get_agent(agent_id)
+
+        cache[agent_id] = create_agent_graph(
+            client=request.app.state.llm_client,
+            agent_skills=agent.skills,
+            system_prompt=agent.build_system_prompt(),
+        )
+
+    return cache[agent_id]