Glonce

CopilotKit ships Channels SDK to bring AI agents into Slack and Teams

agentsThe Glonce editorsHacker News🔥60Glonce rating60Impact50Novelty52Relevance88Surprise38Players42Credibility87
CopilotKit ships Channels SDK to bring AI agents into Slack and Teams

CopilotKit released Channels, an open-source, MIT-licensed SDK that gives an existing AI agent a native presence inside Slack and Microsoft Teams, the chat platforms where a team already works, instead of requiring a separate bot built for each platform. The SDK connects to any AG-UI-compatible agent: the agent keeps its own tools, model and business logic, and Channels lets it understand the ongoing conversation, stream a response, call tools, work with files, render interactive UI inside the chat, and pause to ask a person for approval before taking an action. The project's description says managed connections currently exist for Slack and Microsoft Teams, with more communication platforms planned but not yet named.

The architecture splits the work in two. The developer's agent and application logic run entirely in the developer's own infrastructure. CopilotKit Intelligence, a managed service, handles the connection to the chat platform itself and delivers each conversation turn to the developer's own long-running Channels process, then sends the agent's native UI back into the conversation.

Setup is meant to be driven by a coding agent rather than typed by hand. Running npx copilotkit@latest channels setup installs a skill that walks the coding agent through creating a project, an agent, a managed Channel, a provider app and a long-running runtime; a narrower setup-slack-channel skill can be installed directly for Slack-only setups. The coding agent drives the Slack and Intelligence consoles in the developer's own signed-in browser session and, per the project's own description, is expected to need a browser or computer-use tool to do it. Running a Channel requires Node.js 22 or later and a long-running Node process; the packages @copilotkit/channels and @copilotkit/runtime ship and are versioned together. In the published example, the lifecycle server listens on port 3000 by default and waits up to 30 seconds (30,000 milliseconds) for the channel to report itself online before the process is allowed to serve traffic. The CLI covers the Intelligence side of configuration, while creating the Slack app and installing it into a workspace, plus issuing the project API key, stays a browser step; no CLI flag accepts a credential value, so the bot token and signing secret stay only in the developer's own .env file.

CopilotKit Intelligence can be hosted by CopilotKit itself or self-hosted for enterprise deployments, and the Channels SDK itself is open source under the MIT license, credited in the repository to the organization CopilotKit rather than to a named individual. As a worked example, the project points to OpenTag, described as an open-source, self-hosted on-call triage assistant built with Channels that combines a Python LangGraph agent run over AG-UI with native Slack and Microsoft Teams experiences, file-aware prompts and generative UI, and a human-approval step before it writes to Linear or Notion. The Show HN submission had drawn 94 points and 20 comments roughly 13 hours after posting.

Key facts

  • CopilotKit released Channels, an open-source, MIT-licensed SDK that connects AG-UI-compatible AI agents to Slack and Microsoft Teams with native, interactive chat UI.
  • The developer's agent and application logic run in the developer's own infrastructure; CopilotKit Intelligence, a separate managed service, handles the platform connection and delivers each conversation turn to that process.
  • Running a Channel requires Node.js 22 or later; in the published example the lifecycle server defaults to port 3000 and waits up to 30 seconds for the channel to report itself online.
  • No CLI flag accepts a credential value: the Slack bot token and signing secret are entered by hand and stay only in the developer's local .env file.
  • CopilotKit Intelligence can be hosted by CopilotKit or self-hosted for enterprise use; an example open-source app, OpenTag, pairs Channels with a Python LangGraph agent for on-call triage, including human approval before writes to Linear or Notion.

Why it matters

Teams that already build AI agents typically end up writing separate integration code for every chat platform they want that agent to live in. Channels is CopilotKit's attempt to make that a one-time integration: an agent that already speaks the AG-UI protocol gains a native, interactive presence in Slack and Microsoft Teams, including tool calls, file handling, streamed responses and a pause for human approval, without being rewritten as a platform-specific bot. The project frames this as agents meeting people on the communication surfaces they already use, rather than in a bespoke chat widget.

Who it affects

The audience is developers who already have a working AI agent with its own model, tools and business logic and want to expose it inside Slack or Microsoft Teams without owning the platform integration themselves. It also concerns teams evaluating whether to route that integration through CopilotKit's own hosted Intelligence service or to self-host it as part of an enterprise deployment.

How to use it

The SDK and its example code are open source under the MIT license. Setup is designed to be run by a coding agent: npx copilotkit@latest channels setup installs a skill that walks through creating a project, an agent, a managed Channel, a provider app and a runtime, fetching the current guide from CopilotKit's own site rather than relying on a possibly outdated local copy; a Slack-only path can be installed on its own with npx copilotkit@latest skills install --skill setup-slack-channel -y. Running it requires Node.js 22 or later, the @copilotkit/channels and @copilotkit/runtime packages, and a long-running Node process; the published example server listens on port 3000 by default and waits up to 30 seconds for the channel to come online. The CLI configures the Intelligence side (channels add --adapter slack, channels status), while creating the Slack app and workspace install, and issuing the project API key, happens in the browser; credentials never pass through a CLI flag and stay in the developer's own .env file. CopilotKit Intelligence, the managed piece that brokers the platform connection, can be hosted by CopilotKit or self-hosted for enterprise deployments.

How solid is it

The claims in this story come directly from the project's own repository description rather than from independent testing or a third party, so they describe what the SDK is designed to do rather than an outside verification of it working at scale. As a concrete demonstration, the project points to OpenTag, a separate open-source, self-hosted on-call triage assistant built with Channels that pairs a Python LangGraph agent with native Slack and Microsoft Teams UI, file-aware prompts, generative UI and a human-approval gate before writing to Linear or Notion, described as production-shaped. The Show HN post had reached 94 points and 20 comments about 13 hours after it went up.

Risks and caveats

The source does not give a version number or release date for the SDK, nor any pricing for the hosted CopilotKit Intelligence service, so the cost of the managed path is unclear. Only Slack and Microsoft Teams are supported today; additional channels are described as planned but none are named. The SDK is credited to the organization CopilotKit rather than to a named individual, and running it still requires the developer to operate and maintain their own long-running Node process, since CopilotKit Intelligence only brokers the platform connection rather than hosting the agent itself.