Connecting LLM agents to Slack or Teams usually feels like a

How to actually set it up

If you're trying to move an agent from a local terminal or a basic web UI into a production workspace, the workflow is pretty straightforward. You aren't building the bot from scratch; you're essentially wrapping your existing LLM agent in a layer that the SDK handles.

1. Install the SDK
Start by pulling the package into your project. I usually do this in a clean virtual env to avoid dependency hell.

npm install @channels-sdk/core

2. Configure your Provider
You need to tell the SDK which channel you're targeting. Instead of digging through Slack's massive documentation for event subscriptions, you just define the provider config.

const channel = new ChannelProvider({
  platform: 'slack',
  token: process.env.SLACK_BOT_TOKEN,
  appId: process.env.SLACK_APP_ID
});

3. Bind your Agent logic
This is the part where you plug in your LLM. Whether you're using a custom OpenAI wrapper or a more complex AI workflow, you just pass the message stream to the SDK's handler.

channel.onMessage(async (msg) => {
  const response = await myAgent.process(msg.text);
  await channel.sendMessage(msg.channelId, response);
});

Why this beats manual integration

I've tried building these integrations manually before, and the "gotchas" are everywhere. Slack handles threads differently than Teams, and formatting a simple bold text or a mention can break if you don't follow their specific markdown flavor.

• Unified Payload: You get a consistent JSON object regardless of where the message comes from.
• State Management: It handles the session tracking so your LLM agent doesn't "forget" the conversation context the moment a user sends a second message.
• Deployment Speed: You can move from a prototype to a real-world deployment in a few hours rather than a few days of API debugging.

For anyone doing serious prompt engineering for enterprise tools, the goal is usually to get the agent in front of users as fast as possible. Spending a week on OAuth flows and webhook verification is a waste of time. This SDK lets you focus on the actual agent behavior rather than the plumbing. If you're managing multiple agents, you can just loop through your provider list and broadcast the same agent logic to every channel your team uses.