Channels

Put the agent you already built into Slack or Microsoft Teams. Intelligence owns the connection; your agent stays where it runs.

Overview#

Channels lets the agent you already built answer in Slack or Microsoft Teams. Intelligence owns the connection; your agent stays where it runs.

A Slack channel where a person mentions the agent and asks for a chart. The agent replies in the thread with a native bar chart and a short summary.

Someone mentions the agent in a channel, it runs with the same tools and context it has in your app, and it replies in the thread with native messages, cards, and charts.

What Channels does#

  • One agent, every surface. The agent process you run for your app also serves Slack and Teams. No second agent, no second prompt, no second set of tools.
  • Native replies. Text, buttons, approvals, and charts render as Slack Block Kit or Teams Adaptive Cards. In the screenshot the agent answers a question with a bar chart, not a link.
  • Conversations that persist. Each Slack or Teams conversation shares a stable thread id with the agent run, so Intelligence keeps its history and the run appears in Product Analytics like any other.
  • Identity you control. Your code maps the Slack or Teams user to your app user, so tools and permissions follow the person, and you decide when a run may use User Memory.

How it works#

Intelligence sits between the platform and your agent. It holds the Slack or Teams credentials, receives each message event, and delivers the turn to a long-running Channels SDK process that you run. Your process runs the agent over AG-UI and sends the rendered reply back through Intelligence to the thread.

You never hand platform credentials to your agent, and CopilotKit never runs your agent or your tools. The full message path and the split of responsibilities are on the Channels overview.

Slack is available on cloud-hosted Intelligence. Teams is in controlled availability.

Pick your channel and framework#

Choose where the agent should answer, then the framework it already runs on. The guide you land on starts from that framework and ends with the agent replying in a channel.

Where to go next#

You want toRead
Give the agent application actions and contextTools and context
Map Slack or Teams users to your app usersIdentity and Memory
Send cards, charts, and formatted messagesRich messages and components
Add buttons and approval stepsInteractive messages and approvals
Handle slash commands and emoji reactionsCommands and reactions
Accept files and images from the conversationFiles and multimodal input
See how a channel conversation relates to Rich ThreadsThreads and state
Make workflow state survive a restartPersistence and scaling
Read conversation history or keep a transcriptHistory and transcripts
Run the listener in productionDeploy and operate