opensourceprojects.dev

A broadsheet for software that doesn't ask for your email

Self-hostable workspace for AI coworkers you can actually run yourself
GitHub RepoViews7

Project Description

View on GitHub

OpenBot: A Self-Hostable Workspace for AI Coworkers You Actually Control

You've probably tried a few AI assistants by now. Maybe they were helpful, maybe they were locked behind someone else's cloud, and maybe you couldn't swap out the model or bring in the agents you'd already built. What if you could run the whole thing yourself, pick your own models, and keep your agents on your own infrastructure? That's the pitch behind OpenBot, a modular workspace for AI coworkers from CopilotKit.

What It Does

OpenBot is a self-hostable workspace for AI coworkers. It's built around a simple idea: you choose your models, you bring your agents, and you own your infrastructure. Rather than tying you to a single provider, it gives you a place to run AI assistants that can use tools, browse the web, and hand control back to you when needed.

The project comes in a couple of shapes. For individual use, there's a fully self-contained desktop app that bundles the OpenBot frontend with a Docker-based backend into one application—available as a notarized .dmg for Apple silicon and Intel Macs, plus a signed installer for Windows x64. For teams, there's OpenBot for Business, which adds company sign-in, tool permissions, approvals, and audit, and can be deployed on your own infrastructure. It's built by CopilotKit, the folks behind AG-UI, and you can also run it straight from source or deploy it on Kubernetes.

Why It's Cool

  • It's self-hostable by design. The whole point is that you own the infrastructure. That's a meaningful difference from the usual "sign up and trust us with your data" approach, and it matters if you're working with anything sensitive.

  • Bring your own models and agents. You're not locked into one model provider, and you can connect agents you already run. If you've invested time building agents elsewhere, this gives them a home instead of forcing a rewrite.

  • Approval before action. The demo shows a bot using its browser and tools, answering with interactive results, and then waiting for your approval before it sends. You can take over when it needs you. That human-in-the-loop step is a sensible default for anything that touches the outside world.

  • Getting started doesn't require a terminal. The desktop app is notarized and signed, and the README is explicit that you don't need developer tools. That's a low bar to clear for people who just want to try the thing.

  • A path from laptop to team. You can kick the tires on your own machine and then move to a shared workspace with sign-in, permissions, and audit when you're ready—without switching products.

How to Try It

The fastest route is the desktop app. Grab the build for your platform:

  1. Mac (Apple silicon): OpenBot-mac-arm64.dmg
  2. Mac (Intel): OpenBot-mac-x64.dmg
  3. Windows x64: OpenBot-windows-x64-setup.exe

Install it like any other app—no terminal needed. If you'd rather build it yourself, the README points to Run from source and a Deploy on Kubernetes section for infrastructure-minded folks.

For teams, there's an early access program for OpenBot for Business, which covers company sign-in, tool permissions, approvals, and audit. You can find that link in the README.

Everything lives at github.com/copilotkit/openbot, and there's a docs folder in the repo if you want to dig into the details.

Final Thoughts

OpenBot is aimed at people who want AI assistants without handing over control of their models, agents, or infrastructure—and the desktop app makes it easy enough to evaluate before you commit. The human-approval workflow and the bring-your-own-agents approach are the parts that stand out. If you've been looking for a self-hostable alternative that can grow from a solo experiment into a team workspace, this is worth a look.

Back to Projects
Project ID: openbotLast updated: November 13, 2026 at 02:46 AM