OpenDots is a free, MIT-licensed, self-hostable template for running persistent AI coworkers — named "Dots", each with a role, permissions, its own memory, and optionally its own isolated virtual computer. It comes from CopilotKit, a Seattle company founded in 2023 by Atai Barkai, and it is aimed at developers who are comfortable deploying and configuring a web app themselves and who want a standing AI teammate they run on their own infrastructure instead of an assistant living inside someone else's product.
What it does
The unit of work is a Dot: you create one, give it a name and a job, set what it is allowed to touch, and it keeps its memory between sessions instead of starting from an empty chat every time.
- Each Dot can be given an isolated virtual computer — a browser, a file system and a shell — so it can actually carry a task to the end rather than describe how you would do it.
- You reach the same Dot by text, by a real-time voice call, or by @-mentioning it in Slack, and the conversation context follows you across those channels.
- Finished work lands as editable pages in a "Space": a searchable workspace library with a visual editor and autosave.
- Human-in-the-loop review cards pause risky tool calls — before the Dot saves or sends something — and wait for your approval.
Those are the capabilities the project and its authors describe; nothing here is a benchmark, and there are no performance numbers attached to the release.
How it works
OpenDots is not a standalone product so much as a template on top of CopilotKit's own stack. CopilotKit is an open-source frontend SDK for wiring AI copilots and agentic features into apps (React, Angular, Vue, React Native, Slack, Teams), and it is the author of AG-UI, an open event-based, bidirectional protocol that standardises how a backend agent talks to a user-facing app. CopilotKit describes AG-UI as the user-facing counterpart to Anthropic's MCP: MCP connects an agent to tools and data, AG-UI connects it to the interface a human sits in front of. OpenDots is built end-to-end on CopilotKit plus AG-UI, and it is model-agnostic — it works with any model exposed through an OpenAI-compatible API, not only OpenAI's.
Getting started
The code is free under the MIT licence, but the intelligence is not included. You supply the model: either a paid cloud endpoint with your own API key, or a local OpenAI-API-compatible server running on your own hardware. Using the Slack channel means connecting the deployment to a Slack workspace of your own. The project does not list operating-system, chip, memory or VRAM requirements in the material available here, so treat the hardware question as open — in particular, giving a Dot its own virtual computer means running that sandbox somewhere, and that is capacity you provide.
There is no installer, no release binary and no app-store build: OpenDots is a template you clone from github.com/CopilotKit/OpenDots and deploy yourself. The exact commands are not quoted in the material behind this page, so follow the repository's own setup instructions rather than expecting a one-liner. The release is dated 2 October 2026, which should tell you how much community write-up exists to lean on.
When to use it / when not
The catch is stated plainly by the maker, not discovered by critics: this is alpha software. It is single-owner only, there is no shared editing yet, and the maker describes Slack and voice delegation as still needing more live verification. That makes it a tool for one person's workflow today, not a team deployment.
Use it instead of OpenAI's Dots when the point is control: your infrastructure, your model choice, your data, and source you can read and change. Don't use it if you want something that works for a group on day one, or if you simply want a smarter chat — if you are only asking questions, just asking Claude or ChatGPT costs you less setup and gives you the same answer. OpenDots earns its complexity only when a Dot needs standing memory, a role, and a sandbox to do real work in.
Alternatives
The comparison the project itself makes is OpenAI's "Dots" style of persistent assistant — the same idea of a standing AI teammate with memory and tool access reachable from chat, voice or Slack, but closed, hosted, and tied to one vendor's models. If you want to build your own interface instead of adopting a ready-made one, CopilotKit's SDK and the AG-UI protocol are the layer underneath, and you can use them directly without OpenDots.
Take this one seriously if you are a developer who has already tried to bolt memory and tool access onto a chat assistant and found the plumbing to be the hard part: OpenDots is an open, readable reference implementation of the whole shape — persistent coworkers, approval gates, multi-channel access — from the team that wrote the protocol it runs on. Treat it as an alpha you deploy for yourself, keep the approval cards on, and budget for the model behind it.