CopilotKit/OpenDots

Features

• Persistent Agent Hosting - Hosts long-lived AI agents that keep running over time, each on its own dedicated computer rather than in a short chat.
• Agent Platforms - Hosts long-lived AI agents that keep running over time, each with its own dedicated computer, memory, and permissions instead of a short chat.
• Agent Computers - Provides each agent its own isolated browser, files, and terminal that persist across restarts, with permissions and human takeover.
• Agent Sandboxed Computers - Give each assistant its own browser profile, files, and shell, with human takeover, permission checks, and a record of every action taken.
• Human-in-the-Loop Oversight - Shows a live view of the agent's computer and a review card in the conversation so users can watch and approve actions.
• Tool Execution Approvals - Pauses the conversation on an approval card for write-type tool calls and drafts, continuing only after the user approves or declines.
• Approval Interfaces - Streams agent activity live in chat and shows approval cards before write actions so people stay in control of what agents do.
• Agent Computer Environments - Gives each agent an isolated browser, files, and terminal that persist across restarts, with permission checks, action records, and human takeover.
• Multi-Channel AI Deployments - Carries one AI coworker conversation across written chat, real-time voice calls, and Slack so the same agent follows the user between channels.
• Cross-Channel Agent Conversations - Carries one AI coworker's conversation between written chat, real-time voice calls, and Slack so the agent follows the user.
• Agent Configuration - Defines specialist agents with their own name, role, instructions, and permitted tools so each handles focused work under user control.
• Agent Tool Server Connections - Attaches external tool servers such as email, calendar, and code hosts to an agent, with per-agent tool toggles and approval for writes.
• Voice Agents - Offers a continuous chat timeline plus live voice calls with captions and mute controls while a compute agent handles longer tasks.
• Connected App Tools - Attaches connected apps and external tool servers to each agent with per-tool toggles, requiring owner approval for write actions.
• Conversation Threads - Keeps one persistent conversation thread that follows the user across written chat, voice calls, and Slack.
• Model Context Protocol Servers - Attaches external tool servers to each agent over the Model Context Protocol, with per-agent toggles for exposed tools.
• Specialist Personas - Saves named specialist AI coworkers with their own role instructions and per-assistant permissions for research and memory.
• Web Research Agents - Lets agents discover and read up to five public web sources for a topic or extract supplied URLs, saving each source with its link.
• Nested Page Libraries - Stores working documents as nested pages in shared spaces with a searchable library, grid and list views, and stale-edit protection.
• Shared Human-Agent Workspaces - Provides shared spaces with nested pages where people organize documents and work alongside specialist AI assistants that research, draft, and edit.
• Workspace Spaces and Pages - Lets users create shared spaces, write nested pages, and configure AI assistants locally in the browser before connecting external services.
• Editor Slash Commands - Provides a visual page editor with slash commands for inserting content blocks, plus autosave that rejects stale edits.
• WebRTC Agent Calls - Handles real-time voice calls over WebRTC with time-bounded sessions, captions, hangup control, and a timeline of receipts.
• Slack Integrations - Connects the agent to Slack workspaces, restricted by workspace and user allowlists, so users can mention it and continue in threads.
• Agent Interaction Streams - Streams agent messages, tool calls, and state to the interface so activity and approval cards render inline as work happens.

Star history