fix(moa): keep virtual provider on MoA client

agent/agent_init.py

@@ -765,7 +765,7 @@ def init_agent(
         )
         agent._client_kwargs = {}
         agent.api_key = api_key or "moa-virtual-provider"
-        agent.base_url = base_url or "moa://local"
+        agent.base_url = "moa://local"
         if not agent.quiet_mode:
             print(f"🤖 AI Agent initialized with MoA preset: {agent.model}")
     elif agent.api_mode == "bedrock_converse":

agent/agent_runtime_helpers.py

@@ -1549,7 +1549,14 @@ def switch_model(agent, new_model, new_provider, api_key='', base_url='', api_mo
                 )
 
         # ── Build new client ──
-        if api_mode == "anthropic_messages":
+        if (new_provider or "").strip().lower() == "moa":
+            from agent.moa_loop import MoAClient
+
+            agent.api_key = api_key or "moa-virtual-provider"
+            agent.base_url = "moa://local"
+            agent._client_kwargs = {}
+            agent.client = MoAClient(agent.model or "default")
+        elif api_mode == "anthropic_messages":
             from agent.anthropic_adapter import (
                 build_anthropic_client,
                 resolve_anthropic_token,

agent/chat_completion_helpers.py

@@ -247,6 +247,11 @@ def interruptible_api_call(agent, api_kwargs: dict):
                         invalidate_runtime_client(region)
                     raise
                 result["response"] = normalize_converse_response(raw_response)
+            elif agent.provider == "moa":
+                # MoA is a virtual chat-completions provider backed by the
+                # in-process MoAClient facade. Do not rebuild a request-local
+                # OpenAI client from the virtual runtime metadata.
+                result["response"] = agent.client.chat.completions.create(**api_kwargs)
             else:
                 request_client = _set_request_client(
                     agent._create_request_openai_client(