CopilotKit has released OpenBot 1.0, MarkTechPost reports: an MIT-licensed, self-hostable agent workspace you can run on a laptop or in your own Kubernetes cluster. The pitch is the opposite of a locked-in coworker product. You bring the models, the agents and the machines, and every bot action is supposed to be checked against a policy before it runs.
Each bot gets its own container, with its own browser, logins and files, and it only reaches the tools you grant. Actions go through a governance gateway that evaluates a policy, writes an audit row first, then acts. Deny rules run before allow rules, and a broken rule fails closed. 1.0 adds a desktop app for Mac and Windows, a “teach a bot” flow that turns a recorded task into a reusable skill you review, and the ability for one bot to hand work to specialist bots within limits an admin sets. It speaks the AG-UI protocol, so a list of existing agent frameworks can plug in. Models can come from OpenAI, Claude, Gemini, SpaceXSI’s Grok, or open weights such as Kimi, DeepSeek, Qwen, Nemotron and Mistral.
Personal use is free and can stay on your machine. The business tier is listed at up to $29 per active user per month, with extras like SSO, spend limits and an audit trail.
Review before you install: read the release notes yourself, back up your config, and try new versions somewhere you can afford to break first.
Isolate it: try new software in a disposable VM or on a spare machine first, with a snapshot you can roll back to. Don’t point it at a machine you care about.
My take
Self-hosted, policy-before-action, bring-your-own-model is the shape a lot of people have been asking for. The hard part is still the same: a bot with a browser and logins is a lot of trust, even in a container. If I try this, it’ll be in a throwaway VM, and I’ll read the policy examples before I point it at anything real.
God Bless, and May the Source Be with You!
