refactor: consolidate conversation_history to ConversationHistory class

Remove legacy globals.conversation_history (defaultdict of deques) and
route all callers through utils.conversation_history.ConversationHistory:

- globals.py: remove conversation_history + unused collections imports
- llm.py: remove backward-compat dual-write to legacy system
- api.py: /conversation/{user_id} now reads from ConversationHistory
- actions.py: reset_conversation uses clear_channel()
- figurine_notifier.py: use add_message() instead of buggy setdefault()
- bipolar_mode.py: fix clear_history -> clear_channel (was AttributeError
  silently swallowed by bare except), fix bare except -> except Exception

bot/api.py

@@ -1746,9 +1746,9 @@ def get_autonomous_stats():
 
 @app.get("/conversation/{user_id}")
 def get_conversation(user_id: str):
-    if user_id in globals.conversation_history:
+    """Get conversation history for a user/channel (uses centralized ConversationHistory)."""
-        return {"conversation": list(globals.conversation_history[user_id])}
+    messages = conversation_history.get_recent_messages(user_id)
-    return {"conversation": []}
+    return {"conversation": [{"author": author, "content": content, "is_bot": is_bot} for author, content, is_bot in messages]}
 
 # ========== Figurine DM Subscription APIs ==========
 @app.get("/figurines/subscribers")

bot/commands/actions.py

@@ -4,6 +4,7 @@ import asyncio
 import globals
 from utils.moods import load_mood_description
 from utils.scheduled import send_bedtime_reminder
+from utils.conversation_history import conversation_history
 from utils.logger import get_logger
 
 logger = get_logger('commands')
@@ -32,7 +33,7 @@ def calm_miku() -> str:
 
 
 def reset_conversation(user_id):
-    globals.conversation_history[str(user_id)].clear()
+    conversation_history.clear_channel(str(user_id))
 
 
 async def force_sleep() -> str:

bot/globals.py

@@ -1,14 +1,10 @@
 # globals.py
 import os
-from collections import defaultdict, deque
 import discord
 from apscheduler.schedulers.asyncio import AsyncIOScheduler
 
 scheduler = AsyncIOScheduler()
 
-# Stores last 5 exchanges per user (as deque)
-conversation_history = defaultdict(lambda: deque(maxlen=5))
-
 DISCORD_BOT_TOKEN = os.getenv("DISCORD_BOT_TOKEN")
 
 # Autonomous V2 Debug Mode (set to True to see detailed decision logging)
@@ -26,6 +22,7 @@ VISION_MODEL = os.getenv("VISION_MODEL", "vision")
 EVIL_TEXT_MODEL = os.getenv("EVIL_TEXT_MODEL", "darkidol")  # Uncensored model for evil mode
 JAPANESE_TEXT_MODEL = os.getenv("JAPANESE_TEXT_MODEL", "swallow")  # Llama 3.1 Swallow model for Japanese
 OWNER_USER_ID = int(os.getenv("OWNER_USER_ID", "209381657369772032"))  # Bot owner's Discord user ID for reports
+PREFER_AMD_GPU = os.getenv("PREFER_AMD_GPU", "false").lower() == "true"  # Runtime-overridable via config API
 
 # Cheshire Cat AI integration (Phase 3)
 CHESHIRE_CAT_URL = os.getenv("CHESHIRE_CAT_URL", "http://cheshire-cat:80")

bot/utils/bipolar_mode.py

@@ -1034,8 +1034,8 @@ async def run_argument(channel: discord.TextChannel, client, trigger_context: st
                     
                     # Clean up argument conversation history
                     try:
-                        conversation_history.clear_history(argument_user_id)
+                        conversation_history.clear_channel(argument_user_id)
-                    except:
+                    except Exception:
                         pass  # History cleanup is not critical
                     
                     end_argument(channel_id)

bot/utils/figurine_notifier.py

@@ -5,8 +5,8 @@ from datetime import datetime
 from typing import List, Dict, Any, Tuple
 
 import discord
-import globals
 
+from utils.conversation_history import conversation_history
 from utils.twitter_fetcher import fetch_figurine_tweets_latest
 from utils.image_handling import analyze_image_with_qwen, download_and_encode_image
 from utils.llm import query_llama
@@ -204,15 +204,11 @@ async def send_figurine_dm_to_user(client: discord.Client, user_id: int, tweet:
         # Log the comment message
         dm_logger.log_user_message(user, comment_message, is_bot_message=True)
         
-        # IMPORTANT: Also add to globals.conversation_history for LLM context
+        # Add to conversation history for LLM context (uses centralized ConversationHistory)
         user_id_str = str(user_id)
-        
-        # Add the tweet URL as a "system message" about what Miku just sent (use original URL for context)
         tweet_context = f"[I just sent you this figurine tweet: {tweet_url}]"
-        
+        conversation_history.add_message(channel_id=user_id_str, author_name="Miku", content=tweet_context, is_bot=True)
-        # Add the figurine comment to conversation history 
+        conversation_history.add_message(channel_id=user_id_str, author_name="Miku", content=miku_comment, is_bot=True)
-        # Use empty user prompt since this was initiated by Miku
-        globals.conversation_history.setdefault(user_id_str, []).append((tweet_context, miku_comment))
         
         logger.debug(f"Messages logged to both DM history and conversation context for user {user_id}")
 

bot/utils/gpu_router.py

@@ -37,7 +37,8 @@ MODEL_TO_GPU = {
 }
 
 # Configuration
-PREFER_AMD_GPU = os.getenv("PREFER_AMD_GPU", "false").lower() == "true"
+# PREFER_AMD_GPU lives in globals so the config API can update it at runtime.
+# We read globals.PREFER_AMD_GPU in functions below instead of a frozen local.
 AMD_MODELS_ENABLED = os.getenv("AMD_MODELS_ENABLED", "true").lower() == "true"
 
 
@@ -101,7 +102,7 @@ def get_llama_url_with_load_balancing(
             return globals.LLAMA_URL, "llama3.1"
     
     # AMD enabled - implement load balancing
-    use_amd = prefer_amd or PREFER_AMD_GPU or (random.random() < 0.5)
+    use_amd = prefer_amd or globals.PREFER_AMD_GPU or (random.random() < 0.5)
     
     if task_type == "evil":
         # Evil/uncensored models

bot/utils/llm.py

@@ -475,10 +475,6 @@ Please respond in a way that reflects this emotional tone.{pfp_context}"""
                             is_bot=True
                         )
                     
-                    # Also save to legacy globals for backward compatibility (skip error messages)
-                    if user_prompt and user_prompt.strip() and reply and reply.strip() and reply != "Someone tell Koko-nii there is a problem with my AI.":
-                        globals.conversation_history[user_id].append((user_prompt, reply))
-                    
                     return reply
                 else:
                     error_text = await response.text()