OpenDots: The Open-Source Version of OpenAI's AI Coworkers (Setup Guide)

At DevDay, OpenAI introduced something they called "dots" — always-on AI coworkers that get their own computer, keep working after you close the chat, and bring finished work back when it is ready. Think of them less like a chatbot and more like a background employee who handles tasks on their own machine.

The catch: dots are closed, run only on OpenAI's models, and are limited to certain plans and regions. If you are not on a supported plan or region, or if you simply want to run something like this yourself with your own model and code, you are locked out.

That is the gap CopilotKit stepped in to fill. They released OpenDots — an MIT-licensed, open-source template that lets you self-host your own version. It is built on CopilotKit and AG-UI, and you can bring any OpenAI-compatible model. This article walks through what it is, what it can do, and how to get it running.

What Are OpenAI's "Dots"?

Dots are AI agents that stay active between sessions. Normal AI chat ends when you close the window — dots do not. They have their own computer environment (browser, files, terminal) and can work through tasks over time, then hand results back to you when done.

CopilotKit describes the problem clearly: dots are closed, tied to OpenAI's model stack only, and restricted to specific plans and regions. For developers and teams who want similar functionality without those limits, there was no open alternative — until OpenDots.

OpenDots: The Open-Source Template

OpenDots is a self-hosted template created by CopilotKit. It is available on GitHub at github.com/CopilotKit/OpenDots and had 5,500+ stars at the time of writing (5,574 when checked on October 19, 2026).

A few things to set expectations correctly:

• It is alpha / early development — expect rough edges and active change.
• It is described as "a template, not a hosted product" — there is no managed cloud version you can sign up for; you run it yourself.
• It is a single-owner starting point — designed to get one person or a small team going.
• You can bring any agent via the AG-UI protocol and any OpenAI-compatible model.

What Can You Do with OpenDots?

Here is what the template ships with, based on the README and CopilotKit's announcement:

Specialist Dots. Each Dot is an AI agent you configure with a name, a role description, custom instructions, and a set of tools. For example, you could set up a researcher Dot and a writer Dot — each focused on its own job.

Computer access. A Dot can have its own computer with a browser, files, and a terminal. You can take control of it at any time. This feature is off by default — you need to set up the browser/computer services following the README section called "Computer setup" before it works.

Spaces and Pages. There is a built-in workspace for shared documents — similar to a lightweight team wiki where Dots can write and you can edit.

Real-time voice calls. You can talk to a Dot live, but this requires connecting a speech provider first.

Slack integration. Mention a Dot in a Slack channel and it will respond.

Review-before-save. When a Dot wants to save something, an approval card appears so you can review the change before it is committed.

Scheduled recurring tasks. Set a Dot to run a task on a schedule automatically.

Saved preferences and Automatic Learning. Dots can remember preferences across sessions.

Integrations. MCP (Model Context Protocol) connections and Composio connected apps let Dots interact with external services. Write actions require your explicit approval before they go through.

How to Set It Up

You need Node.js 24 and npm installed. Then follow these steps:

1. Clone and install dependencies

git clone https://github.com/CopilotKit/OpenDots.git
cd OpenDots
npm ci

2. Set up your environment file

cp .env.example .env

3. Start the dev server

npm run dev

Open http://127.0.0.1:5173 in your browser. You can already create Spaces, write pages, and configure Dots at this point — before connecting any services.

4. Connect CopilotKit Intelligence

To actually chat with a Dot, you need a CopilotKit Intelligence project. The hosted route uses the CLI:

npx copilotkit@latest login
npx copilotkit@latest project select

This writes a CPK_INTELLIGENCE_API_KEY value into your .env automatically. There is also a local Docker evaluation option — see the README for that path. Do not run copilotkit onboard inside the OpenDots folder.

5. Add your model keys

Open your .env file and add:

OPENAI_API_KEY=your_key_here
OPENAI_MODEL=your_model_name

6. Restart the dev server

npm run dev

Your Dots can now chat and work.

Tip: If you use an AI coding tool like Claude Code, you can paste the repo URL and ask it to follow the README step by step. That saves you from reading every line manually — but you still need to supply your own API keys.

Things to Know Before You Go Live

Hosted Intelligence = cloud-routed conversations. If you use the hosted CopilotKit Intelligence option (the CLI login path), your conversation data goes through CopilotKit's cloud infrastructure. Your model provider also receives conversation context as part of normal inference. If data privacy is a concern for your use case, read through the README carefully and consider the local Docker evaluation path instead.

Telemetry opt-out. OpenDots collects basic telemetry by default. To disable it, add either of these to your .env:

COPILOTKIT_TELEMETRY_DISABLED=true
# or
DO_NOT_TRACK=1

It is genuinely alpha. Things will break and change. The project is moving fast — if something in this guide is outdated by the time you read it, the README is the authoritative source.

Computer access needs extra setup. The browser/terminal environment for Dots does not work out of the box — you need to follow the "Computer setup" section in the README to enable it.

Why It Matters

OpenDots is not a finished product — but it is a real, working template that shows what AI-native workspaces could look like. The combination of specialist agents, persistent computer access, approval flows, and integrations covers most of what a developer would want when experimenting with autonomous agent workflows.

More importantly, it is open. You can read every line of code, run it locally, and modify it however you need. That is the part dots cannot offer.

Credit where it is due: this is fully CopilotKit's work. The announcement was written by Anmol Baranwal and Eli Berman. If you find OpenDots useful, give the repo a star and follow CopilotKit's updates.