← Back to Discover

OpenDots

An open-source template for building always-on AI agent coworkers, with optional browser and terminal access, document spaces, Slack and voice calls, and approval cards before sensitive actions.

Autonomous agentsAgent frameworks
Visit website ↗

What it does

OpenDots is an open-source starting point for building your own agent workspace, made with CopilotKit. You clone it, define your agents (called Dots), connect your services and adapt the interface and tools. The README is explicit that it is a template rather than a hosted product, that you run it yourself, and that it is in early development. Each Dot has a name, role, instructions and permitted tools, and can be given its own computer with a browser, files and a terminal, with permissions set per Dot and a way for you to take over. Spaces hold working documents that you and the Dots can edit, and a Dot can be asked to show a draft for approval before saving it. Dots can use tools from any MCP server: read-only tools run on their own, and others pause for your approval. Email and calendar access goes through Composio, Slack through a managed Channels SDK connection, and calls pair realtime speech with a separate compute agent. Conversations need CopilotKit Intelligence, either the hosted service, a local Docker evaluation or a licensed self-hosted deployment, plus an OpenAI-compatible model. The README notes where conversation data goes and lists opt-out flags for telemetry. It describes itself as a single-owner starting point without shared editing or invitations.

Capabilities

  • Specialist agents (Dots) with their own roles, instructions and tool permissions
  • Optional per-Dot computer with browser, files and terminal, plus human takeover
  • Spaces for documents with a visual editor and autosave
  • Approval cards before saving drafts or running non-read-only tools
  • MCP tools, Composio-connected apps, Slack and realtime voice calls
  • Scheduled background work

When to use it

  • You want a codebase to start from for an always-on agent workspace that you host yourself
  • You need agents that can browse and run commands only within permissions you set
  • You want a person to approve what an agent saves or runs
  • You are comfortable running Node.js and Docker services and connecting a CopilotKit Intelligence project and a model