Slack AI Council Bot 🤖

A Slack bot that integrates multiple AI models (OpenAI GPT-5.2, Google Gemini 3 Flash Preview, X.AI Grok 3, ByteDance Doubao Seed 1.8) to provide multi-perspective responses in a single thread. Built with Python and the Slack Bolt framework.

✨ Features

Core Functionality

1. Multi-Model Integration 🔌

   • Support for OpenAI (GPT-5.2), Google Gemini (3 Flash Preview), X.AI (Grok 3), and ByteDance (Doubao Seed 1.8)
   • Adapter pattern for easy extensibility (Claude, etc.)
   • Unified interface for all AI models

2. Dynamic Identity Override 🎭

   • Single Slack bot token with multiple AI personas
   • Custom username and emoji for each model
   • Requires chat:write.customize permission

3. Context Isolation & Filtering 🔒

   • Each AI model only sees:
     • User messages
     • Its own previous responses
   • Other AI models' responses are filtered out in Compare mode

4. Async Concurrent Responses ⚡

   • Multiple AI models respond simultaneously
   • Non-blocking execution for fast responses

5. Operation Modes 🎮

   • Compare Mode (default): All models respond concurrently to user questions
   • Debate Mode: Models respond sequentially, seeing each other's responses

6. Interactive Follow-up Questions 💬

   • Each AI response includes a follow-up button
   • Click to open a modal and ask follow-up questions to a specific model
   • Follow-up questions are posted to the thread for visibility
   • Only the selected model responds to follow-up questions

🏗️ Architecture

slack-ai-council/
├── app.py                 # Main Slack bot application
├── llm_manager.py         # LLM adapter pattern implementation
├── context_filter.py      # Message filtering for context isolation
├── mode_manager.py        # Compare vs Debate mode management
├── pyproject.toml         # Project metadata and dependencies (uv)
├── uv.lock                # Locked dependencies (uv)
├── .env.example          # Environment variables template
└── README.md             # This file

Design Patterns

• Adapter Pattern: llm_manager.py provides a unified interface for different AI APIs
• Strategy Pattern: mode_manager.py handles different execution strategies (Compare vs Debate)
• Filter Pattern: context_filter.py filters messages based on model context needs

🚀 Setup

Prerequisites

• Python 3.10 or higher
• Slack workspace with admin access
• API keys for desired AI models:
  • OpenAI API key
  • Google API key (for Gemini)
  • X.AI API key (for Grok)
  • ByteDance API key (for Doubao)

1. Clone the Repository

git clone https://github.com/coolestowl/slack-ai-council.git
cd slack-ai-council

2. Install Dependencies

This project uses uv for fast, reliable dependency management.

# Install uv if you haven't already
pip install uv

# Install project dependencies
uv sync

Alternatively, you can still use pip with the legacy requirements.txt:

pip install -r requirements.txt

3. Create Slack App

1. Go to Slack API Apps
2. Click "Create New App" → "From scratch"
3. Name it "AI Council" and select your workspace

Configure Bot Scopes

In OAuth & Permissions, add these Bot Token Scopes:

• app_mentions:read - Read mentions
• channels:history - Read channel messages
• chat:write - Send messages
• chat:write.customize - Customize message appearance (username/icon)
• im:history - Read DM history
• im:write - Send DMs

Enable Socket Mode

1. Go to Socket Mode and enable it
2. Generate an app-level token with connections:write scope
3. Save the token (starts with xapp-)

Enable Events

1. Go to Event Subscriptions and enable events
2. Subscribe to bot events:
   • app_mention
   • message.im

Install App

1. Go to Install App
2. Install to your workspace
3. Copy the Bot User OAuth Token (starts with xoxb-)

4. Configure Environment Variables

cp .env.example .env

Edit .env and add your tokens and API keys:

# Slack Configuration
SLACK_BOT_TOKEN=xoxb-your-bot-token-here
SLACK_APP_TOKEN=xapp-your-app-token-here

# AI Model API Keys
OPENAI_API_KEY=sk-your-openai-key
GOOGLE_API_KEY=your-google-api-key
XAI_API_KEY=your-xai-key
DOUBAO_API_KEY=your-doubao-api-key

# Optional Configuration
DEFAULT_MODE=compare  # compare or debate

Note: You need at least one AI API key for the bot to work. Missing API keys will be skipped with a warning.

5. Run the Bot

Using uv (recommended):

uv run python app.py

Or directly with Python (if dependencies are already installed):

python app.py

You should see:

✓ Initialized openai adapter
✓ Initialized gemini adapter
✓ Initialized grok adapter
✓ Initialized doubao adapter
==================================================
Slack AI Council Bot Starting
==================================================
Configured AI Models: openai, gemini, grok, doubao
Default Mode: COMPARE
Mode Description: Compare Mode: All AI models respond concurrently
==================================================

💬 Usage

Basic Usage - Mention the Bot

@AI Council What's the best approach to learn machine learning?

All configured AI models will respond in a thread with their perspectives using Compare mode (concurrent responses, default).

Inline Mode Specification

Specify debate mode for sequential AI responses:

@AI Council mode=debate What is the future of artificial intelligence?

Or explicitly use compare mode:

@AI Council mode=compare Explain quantum computing

Modes:

• Compare mode (default): All AI models respond concurrently, each seeing only user messages and their own history
• Debate mode: AI models respond sequentially, seeing all previous responses

Follow-up Questions

Each AI model's response includes a "追问" (Follow-up) button. Click the button to:

1. Open a modal dialog specific to that model
2. Type your follow-up question
3. Submit to get a response from only that model
4. The follow-up question and response appear in the thread

This allows you to have focused conversations with individual models without triggering all models.

🔧 Extending the Bot

Adding New AI Models

Adding a new AI model is simple - just create a new adapter class in llm_manager.py:

class ClaudeAdapter(LLMAdapter):
    adapter_key = "claude"  # Unique identifier for this adapter
    
    def __init__(self):
        super().__init__(
            model_name="claude-3-5-sonnet-20241022",
            username="Claude-3.5",
            icon_emoji=":brain:"
        )
        self.api_key = os.getenv("ANTHROPIC_API_KEY")
        if not self.api_key:
            raise ValueError("ANTHROPIC_API_KEY not found in environment variables")
    
    async def generate_response(self, messages: List[Dict[str, str]]) -> str:
        # Implementation here
        pass

That's it! The adapter will be automatically discovered and registered. No need to modify other files.

Customizing System Prompts

Edit the create_default_system_prompt() function in context_filter.py to customize how each model behaves.

🛠️ Development

Project Structure

• app.py: Main entry point, Slack event handlers
• llm_manager.py: AI model adapters and management
• context_filter.py: Message filtering logic
• mode_manager.py: Operation mode management

Key Functions

• handle_app_mention(): Processes @mentions in channels
• handle_message(): Processes direct messages
• handle_compare_mode(): Concurrent AI responses
• handle_debate_mode(): Sequential AI responses
• filter_messages_for_model(): Context isolation per model

📝 Notes

• The bot uses Socket Mode, so it doesn't require a public URL
• Each AI model only sees user messages and its own history (in Compare mode)
• In Debate mode, models see all previous messages including other AI responses
• Missing API keys are handled gracefully - those models are skipped

🐛 Troubleshooting

"No AI models are configured"

Check that at least one API key is correctly set in your .env file.

Bot doesn't respond

1. Verify Socket Mode is enabled
2. Check that Event Subscriptions are configured
3. Ensure the bot is invited to the channel: /invite @AI Council
4. Check the console for error messages

Custom username not working

Ensure your Slack app has the chat:write.customize permission.

📄 License

See LICENSE file for details.

🤝 Contributing

Contributions welcome! Feel free to:

• Add support for more AI models
• Improve context filtering algorithms
• Enhance debate mode functionality
• Add unit tests

🌟 Future Enhancements

• Advanced debate mode with rounds
• Vote/rating system for best responses
• Per-channel mode configuration
• Response caching
• Streaming responses
• Web dashboard for analytics