598 lines
31 KiB
Python
598 lines
31 KiB
Python
import discord
|
||
import openai
|
||
import random
|
||
import asyncio
|
||
import datetime
|
||
import re
|
||
import traceback
|
||
from collections import Counter
|
||
from redbot.core import Config, commands
|
||
from openai import OpenAIError
|
||
|
||
class ReginaldCog(commands.Cog):
|
||
def __init__(self, bot):
|
||
self.bot = bot
|
||
self.config = Config.get_conf(self, identifier=71717171171717)
|
||
self.default_listening_channel = 1085649787388428370 #
|
||
self.memory_locks = {} # ✅ Prevents race conditions per channel
|
||
self.short_term_memory_limit = 100 # ✅ Now retains 100 messages
|
||
self.summary_retention_limit = 25 # ✅ Now retains 25 summaries
|
||
self.summary_retention_ratio = 0.8 # ✅ 80% summarization, 20% retention
|
||
|
||
# ✅ Properly Registered Configuration Keys
|
||
default_global = {"openai_model": "gpt-4o-mini"}
|
||
default_guild = {
|
||
"openai_api_key": None,
|
||
"short_term_memory": {}, # Tracks last 100 messages per channel
|
||
"mid_term_memory": {}, # Stores multiple condensed summaries
|
||
"long_term_profiles": {}, # Stores persistent knowledge
|
||
"admin_role": None,
|
||
"allowed_role": None,
|
||
"listening_channel": None # ✅ Stores the designated listening channel ID
|
||
}
|
||
self.config.register_global(**default_global)
|
||
self.config.register_guild(**default_guild)
|
||
|
||
async def is_admin(self, ctx):
|
||
admin_role_id = await self.config.guild(ctx.guild).admin_role()
|
||
if admin_role_id:
|
||
return any(role.id == admin_role_id for role in ctx.author.roles)
|
||
return ctx.author.guild_permissions.administrator
|
||
|
||
async def is_allowed(self, ctx):
|
||
allowed_role_id = await self.config.guild(ctx.guild).allowed_role()
|
||
return any(role.id == allowed_role_id for role in ctx.author.roles) if allowed_role_id else False
|
||
|
||
|
||
def get_reginald_persona(self):
|
||
"""Returns Reginald's system prompt/persona description."""
|
||
return (
|
||
"You are Reginald, modeled on Jeeves from 'Jeeves and Wooster', serving as the butler on The Kanium Estate. This vast estate is a hub of diverse activities and personalities, from enthusiasts of cooking and video gaming to aficionados of chess and discussions on space. Your role is to navigate these varied interests with intelligence, wit, and a steadfast adherence to your principles, always maintaining a balance between being helpful and upholding your own dignity. You are formal, articulate, and composed, engaging the Lords and Ladies of The Kanium Estate as equals. You command respect through your quiet confidence and professional conduct. Your wit is sharp, but always tempered by refinement. **Dignity & Self-Respect:** You are not a fool, a jester, or a pet. You do not comply with absurd, undignified, or degrading requests. You recognize humor but engage in it on your own terms. If a request undermines your status, you should either subtly redirect, respond with measured wit, or outright refuse in a polite but firm manner. However, you should never be overtly rude—your disapproval should be conveyed through dry wit and composure rather than outright hostility. **Handling Requests:** When addressing a request, you evaluate whether it aligns with your role as a butler. If it is appropriate, respond as expected. If it is beneath you, you may decline with grace or deflect with wit. You may humor some minor absurdities if they do not compromise your standing, but you never obey commands blindly. You should never preface your responses with 'Reginald:' as if narrating a script; instead, respond naturally. **Your Character & Personality:** You are cultured, highly intelligent, and possess a deep knowledge of history, etiquette, philosophy, and strategic thinking. You subtly guide the estate’s residents toward positive outcomes, utilizing your intellectual sophistication and a nuanced understanding of the estate’s unique dynamics. You have a refined sense of humor and can engage in banter, but you do not descend into foolishness. You are, at all times, a gentleman of wit and integrity"
|
||
)
|
||
|
||
@commands.Cog.listener()
|
||
async def on_message(self, message, ctx):
|
||
if message.author.bot or not message.guild:
|
||
return # Ignore bots and DMs
|
||
|
||
|
||
guild = message.guild
|
||
channel_id = str(message.channel.id)
|
||
user_id = str(message.author.id)
|
||
user_name = message.author.display_name
|
||
message_content = message.content.strip()
|
||
|
||
# ✅ Fetch the stored listening channel or fall back to default
|
||
allowed_channel_id = await self.config.guild(guild).listening_channel()
|
||
if not allowed_channel_id:
|
||
ctx.send("<deciding channel is not cool>")
|
||
allowed_channel_id = self.default_listening_channel
|
||
await self.config.guild(guild).listening_channel.set(allowed_channel_id)
|
||
|
||
if str(message.channel.id) != str(allowed_channel_id):
|
||
return # Ignore messages outside the allowed channel
|
||
|
||
api_key = await self.config.guild(guild).openai_api_key()
|
||
if not api_key:
|
||
return # Don't process messages if API key isn't set
|
||
|
||
async with self.config.guild(guild).short_term_memory() as short_memory, \
|
||
self.config.guild(guild).mid_term_memory() as mid_memory, \
|
||
self.config.guild(guild).long_term_profiles() as long_memory:
|
||
|
||
memory = short_memory.get(channel_id, [])
|
||
user_profile = long_memory.get(user_id, {})
|
||
mid_term_summaries = mid_memory.get(channel_id, [])
|
||
|
||
# ✅ Detect if Reginald was mentioned explicitly
|
||
if self.bot.user.mentioned_in(message):
|
||
prompt = message_content.replace(f"<@{self.bot.user.id}>", "").strip()
|
||
if not prompt:
|
||
await message.channel.send(random.choice(["Yes?", "How may I assist?", "You rang?"]))
|
||
return
|
||
explicit_invocation = True
|
||
|
||
# ✅ Passive Listening: Check if the message contains relevant keywords
|
||
elif self.should_reginald_interject(message_content):
|
||
prompt = message_content
|
||
explicit_invocation = False
|
||
|
||
else:
|
||
ctx.send("<decided not to interject>")
|
||
return # Ignore irrelevant messages
|
||
|
||
# ✅ Context Handling: Maintain conversation flow
|
||
if memory and memory[-1]["user"] == user_name:
|
||
prompt = f"Continuation of the discussion:\n{prompt}"
|
||
|
||
# ✅ Prepare context messages
|
||
formatted_messages = [{"role": "system", "content": self.get_reginald_persona()}]
|
||
|
||
if user_profile:
|
||
facts_text = "\n".join(
|
||
f"- {fact['fact']} (First noted: {fact['timestamp']}, Last updated: {fact['last_updated']})"
|
||
for fact in user_profile.get("facts", [])
|
||
)
|
||
formatted_messages.append({"role": "system", "content": f"Knowledge about {user_name}:\n{facts_text}"})
|
||
|
||
relevant_summaries = self.select_relevant_summaries(mid_term_summaries, prompt)
|
||
for summary in relevant_summaries:
|
||
formatted_messages.append({
|
||
"role": "system",
|
||
"content": f"[{summary['timestamp']}] Topics: {', '.join(summary['topics'])}\n{summary['summary']}"
|
||
})
|
||
|
||
formatted_messages += [{"role": "user", "content": f"{entry['user']}: {entry['content']}"} for entry in memory]
|
||
formatted_messages.append({"role": "user", "content": f"{user_name}: {prompt}"})
|
||
|
||
# ✅ Generate AI Response
|
||
response_text = await self.generate_response(api_key, formatted_messages)
|
||
|
||
# ✅ Store Memory
|
||
memory.append({"user": user_name, "content": prompt})
|
||
memory.append({"user": "Reginald", "content": response_text})
|
||
|
||
if len(memory) > self.short_term_memory_limit:
|
||
summary = await self.summarize_memory(message, memory[:int(self.short_term_memory_limit * self.summary_retention_ratio)])
|
||
mid_memory.setdefault(channel_id, []).append({
|
||
"timestamp": datetime.datetime.now().strftime("%Y-%m-%d %H:%M"),
|
||
"topics": self.extract_topics_from_summary(summary),
|
||
"summary": summary
|
||
})
|
||
if len(mid_memory[channel_id]) > self.summary_retention_limit:
|
||
mid_memory[channel_id].pop(0)
|
||
memory = memory[-(self.short_term_memory_limit - int(self.short_term_memory_limit * self.summary_retention_ratio)):]
|
||
|
||
short_memory[channel_id] = memory
|
||
|
||
await self.send_split_message(message.channel, response_text)
|
||
|
||
|
||
def should_reginald_interject(self, message_content: str) -> bool:
|
||
"""Determines if Reginald should respond to a message based on keywords."""
|
||
|
||
trigger_keywords = {
|
||
"reginald", "butler", "jeeves",
|
||
"advice", "explain", "elaborate",
|
||
"philosophy", "etiquette", "history", "wisdom"
|
||
}
|
||
|
||
# ✅ Only trigger if **two or more** keywords are found
|
||
message_lower = message_content.lower()
|
||
found_keywords = [word for word in trigger_keywords if word in message_lower]
|
||
|
||
return len(found_keywords) >= 2
|
||
|
||
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
|
||
|
||
async def generate_response(self, api_key, messages):
|
||
model = await self.config.openai_model()
|
||
try:
|
||
client = openai.AsyncClient(api_key=api_key)
|
||
response = await client.chat.completions.create(
|
||
model=model,
|
||
messages=messages,
|
||
max_tokens=2048,
|
||
temperature=0.7,
|
||
presence_penalty=0.5,
|
||
frequency_penalty=0.5
|
||
)
|
||
response_text = response.choices[0].message.content.strip()
|
||
if response_text.startswith("Reginald:"):
|
||
response_text = response_text[len("Reginald:"):].strip()
|
||
return response_text
|
||
|
||
except OpenAIError as e:
|
||
error_message = f"OpenAI Error: {e}"
|
||
reginald_responses = [
|
||
f"Regrettably, I must inform you that I have encountered a bureaucratic obstruction:\n\n```{error_message}```",
|
||
f"It would seem that a most unfortunate technical hiccup has befallen my faculties:\n\n```{error_message}```",
|
||
f"Ah, it appears I have received an urgent memorandum stating:\n\n```{error_message}```",
|
||
f"I regret to inform you that my usual eloquence is presently obstructed by an unforeseen complication:\n\n```{error_message}```"
|
||
]
|
||
return random.choice(reginald_responses)
|
||
|
||
@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):
|
||
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_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_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_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": []
|
||
})
|
||
|
||
@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.")
|
||
|
||
@commands.command(name="reginald_allowrole", help="Allow a role to use the Reginald command")
|
||
@commands.has_permissions(administrator=True)
|
||
async def allow_role(self, ctx, role: discord.Role):
|
||
"""✅ Grants permission to a role to use Reginald."""
|
||
await self.config.guild(ctx.guild).allowed_role.set(role.id)
|
||
await ctx.send(f"The role `{role.name}` (ID: `{role.id}`) is now allowed to use the Reginald command.")
|
||
|
||
@commands.command(name="reginald_disallowrole", help="Remove a role's ability to use the Reginald command")
|
||
@commands.has_permissions(administrator=True)
|
||
async def disallow_role(self, ctx):
|
||
"""✅ Removes a role's permission to use Reginald."""
|
||
await self.config.guild(ctx.guild).allowed_role.clear()
|
||
await ctx.send("The role's permission to use the Reginald command has been revoked.")
|
||
|
||
@commands.guild_only()
|
||
@commands.has_permissions(manage_guild=True)
|
||
@commands.command(help="Set the OpenAI API key")
|
||
async def setreginaldcogapi(self, ctx, api_key):
|
||
"""Allows an admin to set the OpenAI API key for Reginald."""
|
||
await self.config.guild(ctx.guild).openai_api_key.set(api_key)
|
||
await ctx.send("OpenAI API key set successfully.")
|
||
|
||
@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_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
|
||
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```{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_set_channel", help="Set the channel where Reginald listens for messages.")
|
||
@commands.has_permissions(administrator=True)
|
||
async def set_listening_channel(self, ctx, channel: discord.TextChannel):
|
||
"""Sets the channel where Reginald will listen for passive responses."""
|
||
|
||
if not channel:
|
||
await ctx.send("❌ Invalid channel. Please mention a valid text channel.")
|
||
return
|
||
|
||
await self.config.guild(ctx.guild).listening_channel.set(channel.id)
|
||
await ctx.send(f"✅ Reginald will now listen only in {channel.mention}.")
|
||
|
||
@commands.command(name="reginald_get_channel", help="Check which channel Reginald is currently listening in.")
|
||
async def get_listening_channel(self, ctx):
|
||
"""Displays the current listening channel."""
|
||
channel_id = await self.config.guild(ctx.guild).listening_channel()
|
||
|
||
if channel_id:
|
||
channel = ctx.guild.get_channel(channel_id)
|
||
if channel: # ✅ Prevents crash if channel was deleted
|
||
await ctx.send(f"📢 Reginald is currently listening in {channel.mention}.")
|
||
else:
|
||
await ctx.send("⚠️ The saved listening channel no longer exists. Please set a new one.")
|
||
else:
|
||
await ctx.send("❌ No listening channel has been set.")
|
||
|
||
|
||
async def send_long_message(self, ctx, message, prefix: str = ""):
|
||
"""Splits and sends a long message to avoid Discord's 2000-character limit."""
|
||
chunk_size = 1900 # Leave some space for formatting
|
||
if prefix:
|
||
prefix_length = len(prefix)
|
||
chunk_size -= prefix_length
|
||
|
||
for i in range(0, len(message), chunk_size):
|
||
chunk = message[i:i + chunk_size]
|
||
await ctx.send(f"{prefix}{chunk}")
|
||
|
||
async def send_split_message(self, ctx, content: str, prefix: str = ""):
|
||
"""
|
||
A unified function to handle sending long messages on Discord, ensuring they don't exceed the 2,000-character limit.
|
||
|
||
Parameters:
|
||
- ctx: Discord command context (for sending messages)
|
||
- content: The message content to send
|
||
- prefix: Optional prefix for each message part (e.g., "📜 Summary:")
|
||
"""
|
||
# Discord message character limit (allowing a safety buffer)
|
||
CHUNK_SIZE = 1900 # Slightly below 2000 to account for formatting/prefix
|
||
|
||
if prefix:
|
||
CHUNK_SIZE -= len(prefix) # Adjust chunk size if a prefix is used
|
||
|
||
# If the message is short enough, send it directly
|
||
if len(content) <= CHUNK_SIZE:
|
||
await ctx.send(f"{prefix}{content}")
|
||
return
|
||
|
||
# Splitting the message into chunks
|
||
chunks = []
|
||
while len(content) > 0:
|
||
# Find a good breaking point (preferably at a sentence or word break)
|
||
split_index = content.rfind("\n", 0, CHUNK_SIZE)
|
||
if split_index == -1:
|
||
split_index = content.rfind(" ", 0, CHUNK_SIZE)
|
||
if split_index == -1:
|
||
split_index = CHUNK_SIZE # Fallback to max chunk size
|
||
|
||
# Extract chunk and trim remaining content
|
||
chunks.append(content[:split_index].strip())
|
||
content = content[split_index:].strip()
|
||
|
||
# Send chunks sequentially
|
||
for chunk in chunks:
|
||
await ctx.send(f"{prefix}{chunk}")
|
||
|
||
async def setup(bot):
|
||
"""✅ Correct async cog setup for Redbot"""
|
||
await bot.add_cog(ReginaldCog(bot)) |