Every engineer who has shipped a Slack bot knows the moment it dies: the user asks something slightly off-script and gets a wall of raw text or nothing at all.

The webhook workaround finally has a real replacement

Connecting an AI agent to a messaging platform today means writing bespoke webhook handlers, flattening rich output into plain text, and repeating that work for every platform. The output is fragile and the maintenance cost compounds every time the agent changes.

An existing agent gets a native interface in two packages

CopilotKit Open Sources Channels SDK takes an agent you already built, connects it over the AG-UI protocol, and renders its output as Block Kit messages in Slack or Adaptive Cards in Microsoft Teams. You install two npm packages, point the SDK at your existing runtime, and the adapter handles the translation to each platform’s native format. Supported agent frameworks include LangGraph, CrewAI, Mastra, Pydantic AI, and Google ADK.

Agent platform engineers are the first to feel the time savings

• Backend engineers shipping internal tools who need agents in Slack without rebuilding orchestration logic for a third time
• Platform teams at mid-size companies who maintain one agent but face requests to put it everywhere employees already work
• AI infrastructure leads evaluating open source options before committing to a proprietary bot platform with lock-in pricing

The SDK is MIT licensed, which changes the procurement conversation entirely for teams that cannot justify a vendor dependency for internal tooling.

Messaging platforms became the de facto AI interface before the tooling caught up

Slack reported over 200,000 third-party app installs monthly as of its last developer report, and most of those integrations are still webhook-to-text pipelines with no real reasoning layer behind them. If the AG-UI protocol gains adoption the way OpenAPI did for REST, the gap between a capable agent and a usable one closes at the transport layer rather than the model layer.

What you can do with it today

• Deploy a LangGraph agent into Slack without rewriting its tool definitions
• Render structured agent output as Adaptive Cards inside Microsoft Teams
• Swap underlying models or orchestration without touching the channel config
• Test against the runtime contract before connecting a live platform

MIT licensed and published on npm. Pricing not listed for the hosted intelligence layer — check our directory.

One real constraint: you are still early

Discord and Google Chat are on the roadmap but not yet available, so teams whose users live in those platforms cannot fully commit yet.

The closest alternative is building directly against the Slack Bolt SDK with a hand-rolled agent loop. For Teams specifically, the Microsoft Bot Framework covers similar ground but requires Azure infrastructure and a steeper configuration surface.

The agent-to-messaging gap is closing faster than most teams planned for

Tools like this are reshaping how engineering teams scope agent deployment — and the teams that catch these shifts early are the ones setting internal standards for everyone else. We cover tools like this every Friday — subscribe here and we’ll send the best ones straight to you.