Agent Barn

Agent Barn is an open source control plane for AI agents, released under the Apache-2.0 license and designed to run on your own Kubernetes. It lets a team run its AI agents from one place instead of managing each one separately. Source code is at https://github.com/aai-labs/agent-barn and guides are at https://agentbarn.dev/guides.

Agents work where teams already talk. You can control your tools from Slack and Teams, and the project also lists Discord and Telegram. Talk to your tools to search Jira, check GitHub activity, update your CRM, and retrieve knowledge through a conversation. Agents can send scheduled reports such as stand-up summaries, project updates, and operational reports, and they can run workflows on autopilot by monitoring events and coordinating multi-step tasks.

Agents are built from versioned templates. A template defines an agent's role, identity, operating instructions, tools, startup behavior, and restrictions in reusable Markdown bundles. You can assign reusable skills and connect services such as Jira, Confluence, GitHub, Bitbucket, and Gmail with encrypted credentials.

Agent Barn also gives you visibility into the whole fleet. You can review conversations, tool executions, logs, lifecycle state, and model costs for every agent your role permits you to access. Management covers organizations, agent access, templates, skills, lifecycle, conversations, tool calls, logs, and costs.

You choose a runtime for each agent, currently OpenClaw or Hermes. Agent Barn renders the pinned configuration, attaches the skills and credentials, and manages the agent's lifecycle.