CopilotKit Intelligence
CopilotKit Intelligence adds AG-UI Streams, User Memories, Automatic Learning, Channels, and Product Analytics to the CopilotKit app you already run.
What is CopilotKit Intelligence?
You want AG-UI Streams, User Memory, Automatic Learning, Channels, and Product Analytics without running the services and storage behind them yourself. CopilotKit Intelligence adds that layer to the CopilotKit app you already have. Your frontend, your agent, and your model stay where they are.
Keep your existing thread system. You do not need to replace your thread storage when you connect Intelligence. Bring your own thread system explains how Intelligence records its own copy while your framework keeps managing the original conversation and agent state.
Connect an existing app in the quickstart, with setup tabs for TypeScript, Python, Go, Ruby, and C#/.NET. TypeScript also works without Intelligence; the other runtime implementations require it.
What Intelligence gives you
Each capability is a page of its own, and every one works with the agent and frontend you already run. Open the one you want to add first.
AG-UI Streams
Complement your existing threads with generative UI, multimodal inputs and file uploads, plus reconnection and catch-up across tabs and devices.
User Memories
Keep facts about a person after the conversation ends.
Automatic Learning
Turn real usage into skills you can review and publish.
Trajectories
Capture browser activity and connect it to agent conversations.
Product Analytics
See what people do with your agent.
Channels
Run the same agent in Slack or Microsoft Teams.
Inspector
Watch threads, learning, and tool calls from your app on localhost.
Managed Slack and Teams Channels can post JSX snapshots as images or combine them with native actions using Render and Carousel.
Server-side runs run an agent in a thread from your server, with no browser attached: a scheduled run, a hand-off between agents, or a resume once a person answers.
Skill delivery combines published Skills from several Learning containers in one agent, with one request for each refresh.
Trajectories record browser activity alongside agent conversations. The Trajectories quickstart connects your app to a local Intelligence project and saves your first Trajectory. These guides are in early access.
Already have agent conversations?
Connect your existing app to Intelligence and add AG-UI streams around your current agent. If you need earlier history, import it once:
- Import LangGraph threads from LangGraph Server / LangSmith deployments or Python FastAPI apps with durable persistence.
- Import Google ADK sessions from supported ADK database session stores or Vertex/Agent Engine session history.
- Import AWS Strands conversations for Python or TypeScript from local session files or S3, plus native AgentCore Memory sessions for Python.
- Import Mastra conversations from a local LibSQL database or a Mastra server.
Import copies history. It does not set up ongoing database replication. New runs through CopilotKit save to Intelligence. If your agent stays connected to its framework storage, those runs also save there. Keep that framework storage connected.
Choose where Intelligence runs
Cloud-hosted and self-hosted use the same app APIs, so you can start on one and move later.
Want to try Intelligence before you choose? Evaluate Intelligence locally in Docker on your Mac with the CLI (preview).
How is this guide?