TrendWhat rose on GitHub, every day at 07:30 KST · 한국어

GitHub · as of November 21, 2026 · View on GitHub →

CopilotKit/CopilotKit

Frontend framework for AI agents

Connect existing AI agents to chat and interfaces for end users

You can connect an already developed AI agent to screens like React or Slack, allowing users to chat, see buttons appear, and receive approval requests. CopilotKit is a frontend stack that delivers an agent's reasoning process and results to the user's screen. Based on the AG-UI protocol, it handles the flow where an agent reads the user's current screen, generates UI, and waits for human approval.

Recent updates: AG-UI 1.0 and Continual Learning

CopilotKit has recently been updated to support AG-UI 1.0 and the A2UI 1.0 spec for declarative generative UI. It also adds a 'Continual Learning' feature that allows agents to learn new skills based on real user data and review and publish them. Slack and Microsoft Teams now include a feature that renders channel code written once using native block kits and adaptive cards.

How agents connect to the screen

The agent handles reasoning and execution, while CopilotKit converts this into a product experience. Events are streamed via AG-UI, allowing you to show the agent's task execution process in real time. Users can see what the agent is currently looking at, call tools to render UI, and experience the process of waiting for human approval before the next step. Throughout this process, the agent logic remains identical across all platforms.

Three forms of generative UI

The way an agent creates UI is divided into three types.

  • Controlled The agent selects and displays from components pre-registered by the developer. It operates based on AG-UI tool calls and state.
  • Declarative The agent describes the UI as a structured spec, and the app renders it. It uses the A2UI spec.
  • Open The agent generates a complete interface that runs in a sandbox. It is based on MCP Apps.

Supported platforms and agent frameworks

React and Next.js are in General Availability (GA) status, and Angular, Vue, and React Native are also supported. Slack and Microsoft Teams are in General Availability on the cloud-hosted Intelligence service. Discord and WhatsApp are available via channel SDK adapters. It is compatible with various agent frameworks such as LangGraph, CrewAI, Mastra, and PydanticAI, and you can also use CopilotKit's built-in agent.

Criteria for production operations

CopilotKit is released under the MIT license, allowing you to start with the open-source version and expand to cloud hosting or self-hosting as needed. The developer plan is free. When self-hosting, you can use the same platform in Kubernetes clusters, VPCs, or air-gapped environments. Before adoption, you should check whether the agent framework you intend to use supports the AG-UI protocol, and if you need external channel integration like Slack or Teams, verify the necessity of the cloud-hosted Intelligence service.

Written by AI from this repository's README on November 1, 2026. For the facts, the original on GitHub is the reference.

By the numbers

Read from GitHub, as of November 15, 2026. None were read on November 21, 2026.

Language
TypeScript
Topics
agent · agent-native · agentic-ai · agents · ai · ai-agent · ai-assistant · assistant
Stars
38,713
Contributors
286
Forks
4,830
Latest release
v1.88.0 · November 13, 2026
Commits
19,937
Last commit
November 13, 2026
Open issues
127
Open pull requests
314
Used by
2,123

Related repositories

View on GitHub → · Homepage