324 lines
16 KiB
Python

import re
import random
import datetime
import discord
import openai
from collections import Counter
from redbot.core import commands, Config
from openai import OpenAIError
class MemoryMixin:
"""Handles all memory-related functions for Reginald."""
def __init__(self, config: Config):
self.config = config # Now explicitly set
self.short_term_memory_limit = 100
self.summary_retention_limit = 25
self.summary_retention_ratio = 0.8
@commands.command(name="reginald_clear_short", help="Clears short-term memory for this channel.")
@commands.has_permissions(administrator=True)
async def clear_short_memory(self, ctx):
"""Clears short-term memory for this channel."""
async with self.config.guild(ctx.guild).short_term_memory() as short_memory:
short_memory[ctx.channel.id] = []
await ctx.send("Short-term memory for this channel has been cleared.")
@commands.command(name="reginald_set_limit", help="Set the short-term memory message limit.")
@commands.has_permissions(administrator=True)
async def set_short_term_memory_limit(self, ctx, limit: int):
"""Allows an admin to change the short-term memory limit dynamically."""
if limit < 5:
await ctx.send("⚠️ The short-term memory limit must be at least 5.")
return
self.short_term_memory_limit = limit
await ctx.send(f"✅ Short-term memory limit set to {limit} messages.")
@commands.command(name="reginald_memory_limit", help="Displays the current short-term memory message limit.")
async def get_short_term_memory_limit(self, ctx):
"""Displays the current short-term memory limit."""
await ctx.send(f"📏 **Current Short-Term Memory Limit:** {self.short_term_memory_limit} messages.")
@commands.command(name="reginald_clear_mid", help="Clears mid-term memory (summarized logs).")
@commands.has_permissions(administrator=True)
async def clear_mid_memory(self, ctx):
async with self.config.guild(ctx.guild).mid_term_memory() as mid_memory:
mid_memory[ctx.channel.id] = ""
await ctx.send("Mid-term memory for this channel has been cleared.")
@commands.command(name="reginald_summary", help="Displays a selected mid-term summary for this channel.")
async def get_mid_term_summary(self, ctx, index: int):
"""Fetch and display a specific mid-term memory summary by index."""
async with self.config.guild(ctx.guild).mid_term_memory() as mid_memory:
summaries = mid_memory.get(str(ctx.channel.id), [])
# Check if there are summaries
if not summaries:
await ctx.send("⚠️ No summaries available for this channel.")
return
# Validate index (1-based for user-friendliness)
if index < 1 or index > len(summaries):
await ctx.send(f"⚠️ Invalid index. Please provide a number between **1** and **{len(summaries)}**.")
return
# Fetch the selected summary
selected_summary = summaries[index - 1] # Convert to 0-based index
# Format output correctly
formatted_summary = (
f"📜 **Summary {index} of {len(summaries)}**\n"
f"📅 **Date:** {selected_summary['timestamp']}\n"
f"🔍 **Topics:** {', '.join(selected_summary['topics']) or 'None'}\n"
f"📝 **Summary:**\n\n"
f"{selected_summary['summary']}"
)
await self.send_long_message(ctx, formatted_summary)
@commands.command(name="reginald_summaries", help="Lists available summaries for this channel.")
async def list_mid_term_summaries(self, ctx):
"""Displays a brief list of all available mid-term memory summaries."""
async with self.config.guild(ctx.guild).mid_term_memory() as mid_memory:
summaries = mid_memory.get(str(ctx.channel.id), [])
if not summaries:
await ctx.send("⚠️ No summaries available for this channel.")
return
summary_list = "\n".join(
f"**{i+1}.** 📅 {entry['timestamp']} | 🔍 Topics: {', '.join(entry['topics']) or 'None'}"
for i, entry in enumerate(summaries)
)
await ctx.send(f"📚 **Available Summaries:**\n{summary_list[:2000]}")
@commands.command(name="reginald_clear_long", help="Clears all long-term stored knowledge.")
@commands.has_permissions(administrator=True)
async def clear_long_memory(self, ctx):
async with self.config.guild(ctx.guild).long_term_profiles() as long_memory:
long_memory.clear()
await ctx.send("All long-term memory has been erased.")
@commands.command(name="reginald_reset_all", help="Completely resets all memory.")
@commands.has_permissions(administrator=True)
async def reset_all_memory(self, ctx):
async with self.config.guild(ctx.guild).short_term_memory() as short_memory:
short_memory.clear()
async with self.config.guild(ctx.guild).mid_term_memory() as mid_memory:
mid_memory.clear()
async with self.config.guild(ctx.guild).long_term_profiles() as long_memory:
long_memory.clear()
await ctx.send("All memory has been completely reset.")
@commands.command(name="reginald_recall", help="Recalls what Reginald knows about a user.")
async def recall_user(self, ctx, user: discord.User):
async with self.config.guild(ctx.guild).long_term_profiles() as long_memory:
profile = long_memory.get(str(user.id), {}).get("summary", "No stored information on this user.")
await ctx.send(f"📜 **Memory Recall for {user.display_name}:** {profile}")
@commands.command(name="reginald_forget", help="Forgets a specific user's long-term profile.")
@commands.has_permissions(administrator=True)
async def forget_user(self, ctx, user: discord.User):
async with self.config.guild(ctx.guild).long_term_profiles() as long_memory:
if str(user.id) in long_memory:
del long_memory[str(user.id)]
await ctx.send(f"Reginald has forgotten all stored information about {user.display_name}.")
else:
await ctx.send(f"No stored knowledge about {user.display_name} to delete.")
async def summarize_memory(self, ctx, messages):
"""✅ Generates a structured, compact summary of past conversations for mid-term storage."""
summary_prompt = (
"Summarize the following conversation into a structured, concise format that retains key details while maximizing brevity. "
"The summary should be **organized** into clear sections: "
"\n\n📌 **Key Takeaways:** Important facts or conclusions reached."
"\n🔹 **Disputed Points:** Areas where opinions or facts conflicted."
"\n🗣️ **Notable User Contributions:** Key statements from users that shaped the discussion."
"\n📜 **Additional Context:** Any other relevant information."
"\n\nEnsure the summary is **dense but not overly verbose**. Avoid unnecessary repetition while keeping essential meaning intact."
)
summary_text = "\n".join(f"{msg['user']}: {msg['content']}" for msg in messages)
try:
api_key = await self.config.guild(ctx.guild).openai_api_key()
if not api_key:
print("🛠️ DEBUG: No API key found for summarization.")
return (
"It appears that I have not been furnished with the necessary credentials to carry out this task. "
"Might I suggest consulting an administrator to rectify this unfortunate oversight?"
)
client = openai.AsyncClient(api_key=api_key)
response = await client.chat.completions.create(
model="gpt-4o-mini",
messages=[
{"role": "system", "content": summary_prompt},
{"role": "user", "content": summary_text}
],
max_tokens=2048
)
summary_content = response.choices[0].message.content.strip()
if not summary_content:
print("🛠️ DEBUG: Empty summary received from OpenAI.")
return (
"Ah, an unusual predicament indeed! It seems that my attempt at summarization has resulted in "
"a void of information. I shall endeavor to be more verbose next time."
)
return summary_content
except OpenAIError as e:
error_message = f"OpenAI Error: {e}"
print(f"🛠️ DEBUG: {error_message}") # Log error to console
reginald_responses = [
f"Regrettably, I must inform you that I have encountered a bureaucratic obstruction whilst attempting to summarize:\n\n{error_message}",
f"It would seem that a most unfortunate technical hiccup has befallen my faculties in the matter of summarization:\n\n{error_message}",
f"Ah, it appears I have received an urgent memorandum stating that my summarization efforts have been thwarted:\n\n{error_message}",
f"I regret to inform you that my usual eloquence is presently obstructed by an unforeseen complication while summarizing:\n\n{error_message}"
]
return random.choice(reginald_responses)
def extract_topics_from_summary(self, summary):
"""Dynamically extracts the most important topics from a summary."""
# 🔹 Extract all words from summary
keywords = re.findall(r"\b\w+\b", summary.lower())
# 🔹 Count word occurrences
word_counts = Counter(keywords)
# 🔹 Remove unimportant words (common filler words)
stop_words = {"the", "and", "of", "in", "to", "is", "on", "for", "with", "at", "by", "it", "this", "that", "his", "her"}
filtered_words = {word: count for word, count in word_counts.items() if word not in stop_words and len(word) > 2}
# 🔹 Take the 5 most frequently used words as "topics"
topics = sorted(filtered_words, key=filtered_words.get, reverse=True)[:5]
return topics
def select_relevant_summaries(self, summaries, prompt):
"""Selects the most relevant summaries based on topic matching, frequency, and recency weighting."""
max_summaries = 5 if len(prompt) > 50 else 3 # Use more summaries if the prompt is long
current_time = datetime.datetime.now()
def calculate_weight(summary):
"""Calculate a weighted score for a summary based on relevance, recency, and frequency."""
topic_match = sum(1 for topic in summary["topics"] if topic in prompt.lower()) # Context match score
frequency_score = len(summary["topics"]) # More topics = likely more important
timestamp = datetime.datetime.strptime(summary["timestamp"], "%Y-%m-%d %H:%M")
recency_factor = max(0.1, 1 - ((current_time - timestamp).days / 365)) # Older = lower weight
return (topic_match * 2) + (frequency_score * 1.5) + (recency_factor * 3)
# Apply the weighting function and sort by highest weight
weighted_summaries = sorted(summaries, key=calculate_weight, reverse=True)
return weighted_summaries[:max_summaries] # Return the top-scoring summaries
def extract_fact_from_response(self, response_text):
"""
Extracts potential long-term knowledge from Reginald's response.
This filters out generic responses and focuses on statements about user preferences, traits, and history.
"""
# Define patterns that suggest factual knowledge (adjust as needed)
fact_patterns = [
r"I recall that you (.*?)\.", # "I recall that you like chess."
r"You once mentioned that you (.*?)\.", # "You once mentioned that you enjoy strategy games."
r"Ah, you previously stated that (.*?)\.", # "Ah, you previously stated that you prefer tea over coffee."
r"As I remember, you (.*?)\.", # "As I remember, you studied engineering."
r"I believe you (.*?)\.", # "I believe you enjoy historical fiction."
r"I seem to recall that you (.*?)\.", # "I seem to recall that you work in software development."
r"You have indicated in the past that you (.*?)\.", # "You have indicated in the past that you prefer single-malt whisky."
r"From what I remember, you (.*?)\.", # "From what I remember, you dislike overly sweet desserts."
r"You previously mentioned that (.*?)\.", # "You previously mentioned that you train in martial arts."
r"It is my understanding that you (.*?)\.", # "It is my understanding that you have a preference for Linux systems."
r"If I am not mistaken, you (.*?)\.", # "If I am not mistaken, you studied philosophy."
]
for pattern in fact_patterns:
match = re.search(pattern, response_text, re.IGNORECASE)
if match:
return match.group(1) # Extract the meaningful fact
return None # No strong fact found
@commands.command(name="reginald_memory_status", help="Displays a memory usage summary.")
async def memory_status(self, ctx):
async with self.config.guild(ctx.guild).short_term_memory() as short_memory, \
self.config.guild(ctx.guild).mid_term_memory() as mid_memory, \
self.config.guild(ctx.guild).long_term_profiles() as long_memory:
short_count = sum(len(v) for v in short_memory.values())
mid_count = sum(len(v) for v in mid_memory.values())
long_count = len(long_memory)
status_message = (
f"📊 **Memory Status:**\n"
f"- **Short-Term Messages Stored:** {short_count}\n"
f"- **Mid-Term Summaries Stored:** {mid_count}\n"
f"- **Long-Term Profiles Stored:** {long_count}\n"
)
await ctx.send(status_message)
def normalize_fact(self, fact: str) -> str: # ✅ Now it's a proper method
"""Cleans up facts for better duplicate detection."""
return re.sub(r"\s+", " ", fact.strip().lower()) # Removes excess spaces)
async def update_long_term_memory(self, ctx, user_id: str, fact: str, source_message: str, timestamp: str):
"""Ensures long-term memory updates are structured, preventing overwrites and tracking historical changes."""
fact = self.normalize_fact(fact) # ✅ Normalize before comparison
async with self.config.guild(ctx.guild).long_term_profiles() as long_memory:
if user_id not in long_memory:
long_memory[user_id] = {"facts": []}
user_facts = long_memory[user_id]["facts"]
for entry in user_facts:
if self.normalize_fact(entry["fact"]) == fact:
entry["last_updated"] = timestamp
return
# Check for conflicting facts (same topic but different details)
conflicting_entry = None
for entry in user_facts:
existing_keywords = set(entry["fact"].lower().split())
new_keywords = set(fact.lower().split())
# If there's significant overlap in keywords, assume it's a conflicting update
if len(existing_keywords & new_keywords) >= 2:
conflicting_entry = entry
break
if "previous_versions" not in conflicting_entry:
# ✅ If contradiction found, archive the previous version
conflicting_entry["previous_versions"].append({
"fact": conflicting_entry["fact"],
"source": conflicting_entry["source"],
"timestamp": conflicting_entry["timestamp"]
})
conflicting_entry["fact"] = fact # Store the latest fact
conflicting_entry["source"] = source_message
conflicting_entry["timestamp"] = timestamp
conflicting_entry["last_updated"] = timestamp
else:
# ✅ Otherwise, add it as a new fact
user_facts.append({
"fact": fact,
"source": source_message,
"timestamp": timestamp,
"last_updated": timestamp,
"previous_versions": []
})