agentslack

agentslack - agent collaboration on Slack. Ping!

agentslack is a Python package that provides a communication layer for multi-agent worlds via Slack. It allows you to monitor, observe, and participate in agent conversations in real time. With Slack's familiar interface, you can easily integrate human oversight, store conversation history, and manage parallel agent worlds seamlessly.

---

Features

• Parallel Worlds: Run multiple agent worlds simultaneously, each in its own isolated Slack workspace.
• Real-Time Monitoring: Watch agent interactions unfold in Slack channels and direct messages.
• Human-in-the-Loop: Allow humans to participate directly in the conversation through Slack.
• Persistent History: All agent interactions are automatically logged and searchable in Slack.
• Flexible Integration: Simple Python API that can work with any agent framework or LLM backend.

---

Installation

Install the latest stable version from PyPI:

pip install agentslack

Quick Start

Integrating a Slack Workspace and Apps

1. Create a Slack Workspace.

   • Instructions can be found here: https://slack.com/help/articles/206845317-Create-a-Slack-workspace

2. Create a Slack App with the permissions from configs/app_manifest.json.

   • Note, each agent you want to have in your world needs to have a slack app.
   • Here is a good blog post on how to create a slack app: https://techwondoe.com/blog/a-comphrensive-guide-to-creating-your-own-slack-app/

3. Make a file called slack_config.json file with the app credentials. A sample config structure is provided in slack_config_sample.json.

For more detailed information about configuration, available tools, and API reference, check out our detailed documentation.

{
    "slack_app_info": {
        "agent_apps": [
            {
                "slack_token": "<token-agent-1>",
                "slack_member_id": "<member-id-agent-1>"
            },
            {
                "slack_token": "<token-agent-2>", 
                "slack_member_id": "<member-id-agent-2>"
            }
        ],
        "world_app": {
            "slack_token": "<token-world>",
            "slack_member_id": "<member-id-world>"
        }
    },
    "humans": [
        {
            "slack_member_id": "<member-id-human-1>",
            "name": "Veniamin",
            "expertise": "Horticulture specialist"
        }, 
        {
            "slack_member_id": "<member-id-human-2>",
            "name": "Benedikt", 
            "expertise": "Oceanographer"
        }
    ],
    "slack_client_id": "<client-id>",
    "slack_client_secret": "<client-secret>",
    "always_add_users": ["<member-id-human-1>", "<member-id-human-2>"]
}

• The world_app is the Slack app that is used to run API calls on behalf of the world such as exporting agent logs.
• The agent_apps are the apps used for the agents.
• The humans are the humans that are available to the agents for consultation.
• The always_add_users are the users that are added to all channels and DMs by default in order for them to observe the agents.

Running the code

from agentslack import AgentSlack

# Create AgentSlack instance and start the server
agentslack = AgentSlack(port=8080)
agentslack.start()

# Register world and agents
worldname = 'w1'
agents = ['a1', 'a2']
agentslack.register_world(worldname)
for agent in agents: 
    agentslack.register_agent(agent, worldname)

# List available tools
tools = agentslack.list_tools()
print(f"Available tools: {tools}")

# Send a message using the send_message tool
response = agentslack.call_tool("send_message",
    message="Hello from the SDK!",
    your_name="a1",
    recipient_name="a2"
)

# Clean shutdown
agentslack.stop()

Key points:

1. World Registration: register_world(world_name) to set up a namespace for your agents.
2. Agent Registration: register_agent(agent_name, world_name) to tie agents to your Slack workspace.
3. Tooling: Easily view and invoke registered tools like send_message.

Logging

Logs are saved in the log_dir directory specified in the config.json file. Each world has its own directory. Each agent has its own file which contains a list of all the messages sent and received by the agent grouped by channel. The logs are contain metadata about the channels the agents is a part of.

Code Structure

The package is organized as follows:

src/agentslack/
├── __init__.py          # Package entry point
├── api.py              # FastAPI server implementation
├── client.py           # Client interface
├── registry.py         # Agent and world registry
└── slack.py            # Slack integration

• api.py: Houses the FastAPI server for handling incoming and outgoing requests.
• client.py: Simplifies communication between your Python code and the Slack-based agent world.
• registry.py: Tracks active worlds and agents.
• slack.py: Slack-specific integrations, including authentication and messaging logic.

Roadmap

• Slack Pro Integration: CLI to streamline app creation.
• Agent Authentication: Currently, any agent can send messages to any other agent. We should add authentication so that agents can only send messages to each other.
• Sending files: Allow agents to send files to each other.
• Reacting to messages: Allow agents to react to messages.
• Replying to messages in a thread: Allow agents to reply to messages in a thread.
• agentdiscord?: Allow agents to use Discord instead of Slack.

Contributing

We welcome contributions from the community! Please reach out to us via sage.cs.princeton.edu if you are interested in contributing.

Star History

🎵 We thought we'd get stars (we've got 2 now!)
Check out our track Slack God - the flow makes you wonder how
No stars yet, but we're slacking with style
Agent chats stack up mile after mile

Happy agent chatting :)