Your AI coworker gets its own computer — you keep the approve button
A self-hostable template for persistent AI agents that chat, take calls, and answer in Slack — each with its own containerized computer and a review gate before saving anything.

What it does — OpenDots is a self-hostable TypeScript template — the README is blunt that it’s a template, not a hosted product — for building persistent AI agents called Dots. Each Dot carries a name, role, instructions, and a permitted toolset, and you reach it through web or mobile chat, voice calls, or Slack mentions. Output lands in Spaces, a document workspace with a visual editor, autosave, and revision checks. A Dot can also be assigned its own computer: a containerized browser, file workspace, and shell supervised by OpenBot, with the browser profile and files persisting across stop/start.
The interesting bit — Voice calls pair realtime speech with a separate compute agent, so the Dot can keep talking while longer work runs; both sides share the same conversation context and tool permissions. And saving is gated: a human-in-the-loop card pauses the tool call for Approve & save or Decline, and retries recover the same saved page instead of duplicating it.
Key highlights
- Spaces: searchable library, nested subpages, slash commands, autosave, and revision checks that stop stale edits overwriting newer content.
- Dot computers via OpenBot: live browser view, human takeover, files, terminal output, and action records, with browser, file, and shell permissions set per Dot.
- Slack through Channels SDK, following the OpenTag pattern with explicit workspace and user allowlists.
- Stack: CopilotKit runtime and React SDK, AG-UI for streamed messages and tool state, TanStack AI for model streaming, Intelligence for durable Threads. MIT, fully self-hostable.
- Sensible default: an unconfigured template executes nothing on your host — computer tools only work once the OpenBot services are connected.
Caveats
- Alpha and single-owner: no shared editing, invitations, or file uploads yet.
- Slack, spoken compute delegation, and Automatic Learning’s cloud schedules and skill delivery are configured locally but still lack connected-service verification.
- The value is the wiring, not novel machinery — it assembles CopilotKit, AG-UI, TanStack AI, OpenBot, and Channels SDK into one reference architecture.
Verdict — If you’re building an agent workspace and want a working reference for persistent, multi-channel agents with sandboxed computers and review gates, this is a serious head start. If you wanted a turnkey product or a team tool, wait — the README says alpha and means it.
Frequently asked
- What is CopilotKit/OpenDots?
- A self-hostable template for persistent AI agents that chat, take calls, and answer in Slack — each with its own containerized computer and a review gate before saving anything.
- Is OpenDots open source?
- Yes — CopilotKit/OpenDots is open source, released under the MIT license.
- What language is OpenDots written in?
- CopilotKit/OpenDots is primarily written in TypeScript.
- How popular is OpenDots?
- CopilotKit/OpenDots has 6.7k stars on GitHub and is currently accelerating.
- Where can I find OpenDots?
- CopilotKit/OpenDots is on GitHub at https://github.com/CopilotKit/OpenDots.