Agent Barn

Agent Barn is an open-source control plane for running and governing AI agents on an organisation's own infrastructure.  Organisations increasingly use AI agents as "coworkers" in team chat tools, where they act on business systems such as code repositories, project trackers and email. Running these agents through external services can make it hard to control access, see what agents did and track what they cost. Agent Barn addresses this by letting organisations run all agents in their own Kubernetes cluster, with oversight features built in.

Users define each agent from a template, choose its language model and connect it to Slack, Microsoft Teams, Telegram or Discord. Ready-made templates include a pull request reviewer, a documentation agent and a Jira task helper. Agents gain skills, such as access to GitHub, Jira or Confluence, only once the matching credential is attached. Credentials are encrypted and stored in the organisation's own database.

Several features support governance. Organisation roles and separate per-agent access roles control who can manage each agent, with each organisation's data kept separate. Every agent has its own model key, so token use and spending can be attributed to individual agents and organisations. Conversation history, a log of every tool call and a trail of system events record what each agent did. One runtime supports a per-agent setting requiring command approval. Agent Barn is written in Python and TypeScript and released under the Apache 2.0 licence.