Getting any agent into messaging platforms like Slack is complex. You create a Slack app, handle its events, format messages to the platform's form...
For further actions, you may consider blocking this person and/or reporting abuse
Can I self-host and skip CopilotKit Intelligence entirely by setting up my own webhook/tunnel to Slack/Teams instead of the managed path? If so, is there documentation for that direct approach? Also curious — will Intelligence remain free, or is pricing planned once usage scales up? Would help a lot for evaluating production costs.
yeah let me answer all of it. CopilotKit Intelligence is itself self-hostable so you can run the whole thing in your infra.
Here is the direct adapters docs, my bad I probably should have linked it somewhere. It supports Slack, Microsoft Teams, WhatsApp, Discord, Telegram right now -- so you can move the provider connection into your process but runtime still uses Intelligence, so you will need the Intelligence API key either way (free).
The pricing for channels is on copilotkit.ai/pricing. The free tier is genuinely enough for testing and evaluation. hope this helps. feel free to ask if you have any more questions!!
pls go to playground page: copilotkit.ai/try-channels and you will get public channel invite once you drop your email, just join the workspace and try the agent live...
Hot take: I don't want AI anywhere near my personal chats. iMessage, WhatsApp, my group chat, keep it out for sake. At work? heck yeah. That's where I want an agent in the thread, because the context is already there and the whole team sees it.
AI companies are stuffing it everywhere. Look at the difference in the last four years, can't imagine what the next four look like. I used to believe AI agents were f useless. Can't say that anymore
same I don't want AI in WhatsApp or my community of real peopl. but if it helps with the work stuff, then it's great. I have been trying it out within the company and it's super useful for some specific situations!
People calling this "just a Slack integration" have it backwards. The Slack part is the easy bit, anyone can echo a model into a channel. But the agent that can actually act in a shared work channel is much more dangerous than useful until the permissions and approvals are airtight. Most teams shipping this says it as AI coworkers right now - and are one bad tool-call away from an agent confidently doing the wrong thing in front of everyone. The hard problem was never getting it into Slack. It's making it completely safe to leave there.
yeah authrorization and approvals matters a lot since I don't want the agent to do some stupid stuff. I also believe that per-scope is very useful too - like eng team's agent shouldn't have the same tools or permissions as sales.. and one channel's context shouldn't leak into another.
And if you layer self-learning on top of that, the agent gets better the more it gets used. lots of exciting possibilities.
Detailed tutorial. how's this different from Vercel's chat SDK, and somehow all of this feels like just a Slack integration at the end of the day.
I knew someone would ask this for sure. In my opinion, it's a different bet on the same problem.
First, ours is built on AG-UI so the agent side is decoupled by protocol, you can go use LangGraph, Mastra, ADK, Claude Agent SDK, Pydantic, Agno, CrewAI or your own loop without touching the channel code -- which itself is a big deal. You are not tied to one AI SDK.
Second, there's CopilotKit Intelligence that owns delivery, retries, ordering, and identity for you, so you're not running the socket/webhook reliability layer yourself, though you can go self-hosted if you want.
there are genuine differences like our agent will be able to self-learn over time (which is already possible with intelligence with other surfaces).. personally I haven't used Chat SDK so I can't say what else do they support
pls go to copilotkit.ai/try-channels and you will get public channel invite once you drop your email, just try the agent live...
it may feel like slack integration but some of those features are really good. and once you add self-learning and other stuff that compounds with context - it will be something much more
oh nice this is really great, let me try it out. what's the catch in running in production?
thank you for trying! production gives you more visibility for long running agents. there is no catch lol.
it holds an outbound connection to Intelligence, so you can deploy it like a queue consumer or a websocket worker (container/VM that stays up), with a health check and clean shutdown. Deploy and operate covers the worker setup.
You can bring your own store: Redis, Postgres, etc. It's a pluggable interface so you can back it with anything you already run.
Overall Managed Intelligence handles the annoying platform side (delivery, retries, ordering).
let me know if you have any feedback once you try it out :)