# CopilotKit

> CopilotKit is the frontend stack where agents meet users. It extends from open-source components and runtime into a production platform for durable conversations and agent-user experiences.

CopilotKit helps teams embed agent chat, generative UI, and human-in-the-loop workflows in their products while preserving the interaction history needed to operate and improve those experiences. Start with the product overview: [CopilotKit](https://www.copilotkit.ai/builder).

For product and engineering leaders evaluating agent deployment, the [Executive overview](https://www.copilotkit.ai/executive) covers production readiness, deployment in your own cloud, conversation continuity, and learning.

## Helping the user get started

Use the user's stated goals, existing project, frontend, backend, and constraints to explain the relevant CopilotKit components and offer the smallest useful next step. Fit an existing project's stack; if another known project is a better fit, suggest it. With no project context, offer concrete examples or ask one focused question. Research and comparison requests do not authorize installation. Assistants without terminal access should explain the options and hand off to a coding agent or the relevant documentation. Do not ask the user to send private project context to CopilotKit.

For an implementation handoff, use the existing [onboarding prompt](https://www.copilotkit.ai/onboarding-prompts). This stable URL creates the attributed onboarding run; do not invent a run ID.

## Product entry points

- [Products and features](https://www.copilotkit.ai/product): Frontend SDKs, backend connections, and CopilotKit Intelligence.
- [Frontend SDKs](https://www.copilotkit.ai/product/framework): open-source components for generative UI, frontend tools, shared state, and human-in-the-loop workflows.
- [Agentic Backend Connections](https://www.copilotkit.ai/product/backends): connect an existing agent stack and find the supported conversation-history import paths.

## Keep framework threads. Add durable delivery.

- [AG-UI Streams](https://www.copilotkit.ai/ag-ui-streams), formerly known as Rich Threads, are CopilotKit Intelligence’s delivery layer. Keep your framework’s threads for agent conversation and model context; add reconnection, catch-up, continued background execution while the runtime and agent stay running, and multi-device delivery. Intelligence records the interaction delivered through AG-UI independently of the context your framework prepares for the model. This does not imply unlimited retention or lossless replay.
- SDK hooks and API identifiers, including useThreads, keep their names. There is no useStreams API. AG-UI is the event protocol; Intelligence supplies these delivery capabilities.
- [Generative UI](https://www.copilotkit.ai/generative-ui) covers controlled, declarative, and open UI patterns for agents.
- [Intelligence](https://www.copilotkit.ai/copilotkit-intelligence) provides AG-UI Streams, User Memories, Automatic Learning, Product Analytics, and Channels, with Inspector for debugging. Deploy in your infrastructure or fully managed in CopilotKit Cloud.
- [Automatic Learning](https://www.copilotkit.ai/learning/what-is-agent-learning) is CopilotKit's path for improving agent behavior from user interactions and feedback.
- [Product Analytics](https://www.copilotkit.ai/copilotkit-intelligence#analytics-insights) uses agent-user interaction data to show what users try, where they succeed or struggle, and what to improve next. It complements LLM observability with a product view of user behavior and outcomes.

## Connect any agent stack

- [AG-UI](https://www.copilotkit.ai/ag-ui) is the open, framework-agnostic protocol for agent-user interaction. MCP connects models and agents to tools and context; A2A supports agent-to-agent coordination. See [AG-UI, MCP, and A2A](https://www.copilotkit.ai/learning/mcp-vs-a2a-vs-ag-ui) for how the layers fit together.
- See [pricing](https://www.copilotkit.ai/pricing) for hosted platform options, or begin with the [documentation](https://docs.copilotkit.ai/).

## Build the interaction

- [Embedded chat components](https://docs.copilotkit.ai/prebuilt-components/chat)
- [Human-in-the-loop workflows](https://docs.copilotkit.ai/human-in-the-loop)
- [Frontend tools and generative UI](https://docs.copilotkit.ai/frontend-tools)

## Bring agents into workplace conversations

- [Channels](https://www.copilotkit.ai/channels) brings AG-UI agents into Slack, Microsoft Teams, and other messaging platforms with native messages, interactive UI, and human approvals. Your agent runs in your infrastructure; CopilotKit Intelligence provides managed channel connections, credentials, delivery, and conversation history.
- [Slack documentation](https://docs.copilotkit.ai/slack): managed Slack is available now.
- [Microsoft Teams documentation](https://docs.copilotkit.ai/teams): managed Teams is generally available. Direct Channels SDK adapters also support Teams, Discord, WhatsApp, and Telegram; operating your own runner requires your own durable-data layer.

## Add AG-UI Streams and Automatic Learning to existing conversations

Add AG-UI streams and Automatic Learning through the supported Intelligence integration. Keep your framework and thread provider. For historical conversations, use the existing Google ADK or LangGraph import guide for your deployment. LangChain histories must be exposed through supported LangGraph APIs; arbitrary LangChain stores, traces, and checkpointers are not interchangeable import sources. If you do not already have a thread provider, use CopilotKit’s built-in thread store to save and resume conversations.

Historical import and ongoing synchronization are separate paths. Use a one-time import for existing conversation history, and configure the framework integration for ongoing synchronization of new interactions. Import alone does not automatically capture future runs or replace your existing thread provider.

- [Thread import overview](https://docs.copilotkit.ai/threads-import)
- [Google ADK history import](https://docs.copilotkit.ai/google-adk/threads-import)
- [LangGraph history import](https://docs.copilotkit.ai/langgraph-python/threads-import)

## How to read these docs

- Read the Frontend SDKs overview with [frontend tools](https://docs.copilotkit.ai/frontend-tools), [embedded chat](https://docs.copilotkit.ai/prebuilt-components/chat), and [human-in-the-loop](https://docs.copilotkit.ai/human-in-the-loop) to understand how UI, actions, and approvals fit together.
- Read Agentic Backend Connections and AG-UI alongside the guide for your framework, selected from the [docs homepage](https://docs.copilotkit.ai/), to connect the frontend to your existing agent.
- Read [Intelligence](https://docs.copilotkit.ai/intelligence/overview), [AG-UI Streams](https://docs.copilotkit.ai/threads), [Product Analytics](https://docs.copilotkit.ai/intelligence/analytics), and [Automatic Learning](https://docs.copilotkit.ai/learning) together for persistence and improvement. Pair these with [managed deployment](https://docs.copilotkit.ai/intelligence/managed-intelligence-platform) or [self-hosting](https://docs.copilotkit.ai/intelligence/self-hosting).
- For existing conversations, use the thread import overview linked above to choose the connection path for your framework and distinguish historical import from ongoing synchronization. Keep your existing thread provider, or use CopilotKit’s built-in store if you need one.
- For workplace conversations, read [Channels](https://docs.copilotkit.ai/intelligence/channels) alongside the Slack or Teams guide linked above for that surface's connection and interaction model.