Add MiniMax LLM integration and local .env support

- Add .env to .gitignore (API keys stay local)
- Add LLM client with MiniMax and OpenAI support
- Update config to load from environment variables
- Wire up Architect agent to actually call the LLM
- Add MiniMax API key to local .env file

opus_orchestrator/agents/base.py

@@ -1,11 +1,12 @@
 """Base agent class for Opus Orchestrator."""
 
 from abc import ABC, abstractmethod
-from typing import Any, Generic, TypeVar
+from typing import Any, Generic, Optional, TypeVar
 
 from pydantic import BaseModel
 
 from opus_orchestrator.config import AgentConfig, get_config
+from opus_orchestrator.utils.llm import LLMClient, get_llm_client
 
 
 T = TypeVar("T", bound=BaseModel)
@@ -23,9 +24,6 @@ class AgentResponse(BaseModel):
         arbitrary_types_allowed = True
 
 
-from typing import Optional
-
-
 class BaseAgent(ABC, Generic[T]):
     """Base class for all Opus agents.
 
@@ -49,6 +47,14 @@ class BaseAgent(ABC, Generic[T]):
         self.system_prompt = system_prompt
         self.output_schema = output_schema
         self.config = config or get_config().agent
+        self._llm_client: Optional[LLMClient] = None
+
+    @property
+    def llm_client(self) -> LLMClient:
+        """Get or create LLM client."""
+        if self._llm_client is None:
+            self._llm_client = get_llm_client()
+        return self._llm_client
 
     @abstractmethod
     async def execute(self, input_data: Any, context: dict[str, Any]) -> AgentResponse:
@@ -63,6 +69,31 @@ class BaseAgent(ABC, Generic[T]):
         """
         pass
 
+    async def call_llm(
+        self,
+        system_prompt: str,
+        user_prompt: str,
+        temperature: Optional[float] = None,
+    ) -> str:
+        """Call the LLM with prompts.
+        
+        Args:
+            system_prompt: System prompt
+            user_prompt: User prompt
+            temperature: Optional temperature override
+            
+        Returns:
+            Generated text
+        """
+        temp = temperature if temperature is not None else self.config.temperature
+        
+        return await self.llm_client.complete(
+            system_prompt=system_prompt,
+            user_prompt=user_prompt,
+            temperature=temp,
+            max_tokens=self.config.max_tokens,
+        )
+
     def build_system_prompt(self, context: dict[str, Any]) -> str:
         """Build the full system prompt with context.
 
@@ -104,3 +135,9 @@ class BaseAgent(ABC, Generic[T]):
 
 Please complete this task following the methodology specified in your system prompt.
 """
+
+    async def cleanup(self):
+        """Clean up resources."""
+        if self._llm_client:
+            await self._llm_client.close()
+            self._llm_client = None

opus_orchestrator/agents/fiction/architect.py

@@ -84,8 +84,6 @@ class ArchitectAgent(BaseAgent):
         Returns:
             AgentResponse with BookBlueprint
         """
-        # This is a placeholder - actual implementation would call the LLM
-        # For now, we'll structure the prompt
         raw_content = input_data.get("raw_content", "")
         intent = input_data.get("intent", {})
         genre = intent.get("genre", "general")
@@ -107,23 +105,34 @@ class ArchitectAgent(BaseAgent):
 
 Generate a complete story blueprint following the Architect's methodology.
 Include all sections specified in your system prompt.
+
+Be specific and detailed. The blueprint should be comprehensive enough that another agent could write each chapter from it.
 """
 
-        # In actual implementation, this would call the LLM
-        # For now, return a structured response
-        return AgentResponse(
-            success=True,
-            output={
-                "status": "blueprint_generated",
-                "message": "Blueprint generation would be executed here with LLM",
-            },
-            metadata={
-                "role": "Architect",
-                "input_word_count": len(raw_content.split()),
-                "target_word_count": target_word_count,
-                "genre": genre,
-            },
-        )
+        try:
+            # Call the LLM
+            result = await self.call_llm(
+                system_prompt=self.build_system_prompt(context),
+                user_prompt=user_prompt,
+            )
+
+            return AgentResponse(
+                success=True,
+                output=result,
+                metadata={
+                    "role": "Architect",
+                    "input_word_count": len(raw_content.split()),
+                    "target_word_count": target_word_count,
+                    "genre": genre,
+                },
+            )
+        except Exception as e:
+            return AgentResponse(
+                success=False,
+                output=None,
+                error=str(e),
+                metadata={"role": "Architect"},
+            )
 
     async def expand_chapter(
         self,
@@ -157,13 +166,31 @@ Include all sections specified in your system prompt.
 
 Expand this chapter beat into a detailed scene specification following
 Template B from the Fiction Fortress methodology.
+
+Include:
+1. Opening beat - how the scene opens
+2. Conflict beat - what escalates tension
+3. Turn beat - what changes the situation
+4. Ending beat - what hook or change ends the scene
+
+Be specific about character motivations, dialogue objectives, and emotional progression.
 """
 
-        return AgentResponse(
-            success=True,
-            output={
-                "status": "chapter_expanded",
-                "chapter_number": chapter.chapter_number,
-            },
-            metadata={"role": "Architect", "task": "chapter_expansion"},
-        )
+        try:
+            result = await self.call_llm(
+                system_prompt=self.build_system_prompt(context),
+                user_prompt=user_prompt,
+            )
+
+            return AgentResponse(
+                success=True,
+                output=result,
+                metadata={"role": "Architect", "task": "chapter_expansion"},
+            )
+        except Exception as e:
+            return AgentResponse(
+                success=False,
+                output=None,
+                error=str(e),
+                metadata={"role": "Architect"},
+            )