Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free
Promoted
Launching today
Create, configure, run, and monitor AI agents across Slack, Microsoft Teams, and Telegram from one self-hosted control plane.Agent Barn is an open-source control plane for AI agent fleets. Agents are Markdown files you fork, pin and roll back. Each runs in its own pod with its own credentials. Inference goes through OpenRouter; LiteLLM attributes every token per agent, per model. Helm-install on your own Kubernetes, or clone it and the whole stack comes up on your laptop. Apache 2.0.
We just published our open-source control plane for AI agents. One command, six steps, and you have a working agent in your own Kubernetes cluster, talking to your tools, on your infrastructure. It runs OpenClaw and Hermes under the hood so if you wanted an easy and secure way to get an agent to every employee, now it's a great time to do that.
𝗪𝗵𝘆 𝘄𝗲 𝗯𝘂𝗶𝗹𝘁 𝗶𝘁
Most teams we work with have one hard requirement: software runs on their machines, not in someone else's cloud. The moment a sales call turns to deployment, that requirement kills half the deals.
Agent Barn is our answer. It is open source under Apache 2.0, self-hostable with Helm and PostgreSQL, and built to be the thing your IT team is willing to approve.
When a manufacturing client asks "can we see the source, can we run it on our hardware, can we audit it," the answer is now "yes, today, here is the repo."
𝗪𝗵𝗮𝘁 𝗶𝘀 𝗶𝗻 𝘁𝗵𝗲 𝗯𝗼𝘅
• Agents in Slack, Microsoft Teams, Telegram, and Discord, plus a chat surface in the web UI.
• Per-agent model and cost attribution through an LLM proxy that lives in your namespace. No opaque monthly bill.
• Versioned agent templates. Skills with encrypted credentials. A real audit trail of conversations, tool calls, and lifecycle events.
• A control plane for orgs, agent access, lifecycle, costs, and observability. One place, not five.
• Integrated tools and skills that work with your stack out-of-the-box.
𝗪𝗵𝘆 𝗻𝗼𝘄
Two years of agent tooling taught us that the platform matters. People do not want another model picker. They want a way to give an agent a job, give it the right tools, and watch it do the work.
This is the thing we wanted to build for the last twelve months. It is the thing we wish existed when we started. It is the foundation for every paid agent engagement we sell.
𝗪𝗵𝗮𝘁 𝗶𝘀 𝗻𝗲𝘅𝘁
We will be working with our manufacturings client this quarter. Even though the Agent Barn is open source, manufacturing requires custom implementations, integrations, and vertical-specific templates to make it work.
We are aiming for 100 GitHub stars in the first month. If you have an opinion on agent infrastructure, if you have a use case that needs a real platform under it - give it a star, open an issue, join the Discord. That signal is what we will use to decide how fast to push the roadmap.
in case you want to contribute:
Repo: https://github.com/aai-labs/agent-barn (Stars would be awesome :))
Docs: https://agentbarn.dev/guides
Mailwarm
Honestly, this is pretty cool. The self-hosted part especially caught my attention. I can see why that would matter a lot for teams with strict infrastructure requirements. Congrats!!