Summarizer Cog Documentation¶

Overview¶

The Summarizer cog provides AI-powered conversation summarization using LangChain agents. It allows users to summarize recent chat history in a Discord channel, extracting key themes, decisions, and action items while maintaining links back to the original messages.

Features¶

AI-Powered Summarization: Uses advanced language models (via LangChain) to synthesize conversation history.
Source Mapping: Automatically maps AI-generated references back to original Discord messages with clickable links.
Configurable Scope: Users can specify how many messages to analyze.
Custom Personas: Support for custom AI personas (e.g., "Professional Meeting Recorder", "Witty Critic").
Robust Text Splitting: Automatically handles long summaries by splitting them into multiple Discord embeds.
Dual Limits: Enforces both message count and character count limits to prevent token overflow.

Commands¶

`/summarize`¶

Summarizes the conversation history of the current channel.

Parameters: - limit (int, optional, default: 100): The maximum number of messages to search back in history. - persona (string, optional): Set the AI's summary persona (e.g., "A professional meeting recorder"). - only_me (boolean, optional, default: False): If True, the summary will only be visible to you (ephemeral message).

Behavior: 1. Fetches channel history up to the specified limit. 2. Filters out non-text messages and prioritizes human messages for analysis. 3. Formats messages with unique IDs ([MSG-1], [MSG-2], etc.) and timestamps. 4. Calls the AI model with a specialized system prompt. 5. Post-processes the AI output to replace [MSG-ID] tags with clickable [[Source]] links. 6. Sends the summary as one or more rich embeds.

Technical Implementation¶

Class Structure¶

class SummarizerCog(commands.Cog):
    def __init__(self, bot):
        self.MAX_CHAR_COUNT = 15000  # Character limit for input context
        self.EMBED_DESC_LIMIT = 4000  # Character limit for individual embed descriptions

Process Flow¶

1. Message Collection¶

The cog iterates through the channel history in reverse chronological order. It collects human messages until either the limit is reached or the MAX_CHAR_COUNT (15,000 characters) is exceeded. Bot messages are included as context but marked differently.

2. Prompt Engineering¶

The system prompt instructs the AI to: - Identify core themes, questions, and decisions. - Extract action items. - Ignore small talk. - Use a bulleted list format. - Crucial: Attach the source [MSG-ID] to every summary point.

3. AI Invocation¶

Uses langchain.agents.create_agent with the init_chat_model. It includes a ModelCallLimitMiddleware to ensure exactly one model call per request, preventing runaway agent loops.

4. Post-Processing & Linking¶

The raw AI response is processed using regex to find [MSG-ID] patterns. These are cross-referenced with a source_mapping dictionary containing the jump_url of the original Discord messages, resulting in clickable links in the final embed.

5. Output Management¶

If the generated summary exceeds Discord's embed description limit (4096 characters), _split_text_robustly divides the content into logical chunks and sends them as sequential embeds.

Dependencies¶

discord.py: Discord API interaction
langchain: AI agent and message management
langchain_core: HumanMessage and AIMessage schemas
ModelManager: Centralized LLM configuration
func.report_error: Standardized error reporting

Usage Examples¶

Basic Summary¶

/summarize limit:50

Summarizes the last 50 messages using the default professional persona.

Professional Persona¶

/summarize limit:100 persona:"A strict military officer" only_me:True

Summarizes the last 100 messages with a military tone, visible only to the user.

cogs/summarizer.py: Main implementation
llm/model_manager.py: Model selection and fallback logic
function/func.py: Utility functions and error reporting