Copilotkit

React SDK for embedding AI copilots and agentic assistants into web applications. CopilotKit provider wraps the React tree and manages chat state. useCopilotChat hook for chat UI rendering. useCopilotAction for defining agent-executable actions. useCopilotReadable for injecting real-time app context into the LLM. CoAgent for integration with LangGraph multi-agent workflows. Confidence is docs-derived.

React hooks API: useCopilotChat() for chat state and send functions; useCopilotAction(name, { description, parameters, handler }) to register callable actions; useCopilotReadable(description, value) for context injection; CopilotPopup and CopilotSidebar prebuilt UI components; CopilotTextarea for AI-augmented text inputs; REST backend via @copilotkit/runtime Express/Next.js adapter; CoAgent hooks for LangGraph agent state streaming.

CopilotKit Cloud API key passed to <CopilotKit publicApiKey={...}>; self-hosted runtime uses LLM provider keys (OpenAI, Anthropic, Groq) on the server-side runtime — never exposed to browser; useCopilotAction handlers run client-side within the app's security context; no separate auth layer for actions — access control is the host app's responsibility.

Chat state managed in React context — no external round-trip for UI state; action handlers are synchronous or async and return structured results to the LLM; streaming via SSE (Server-Sent Events) for token-by-token chat responses; LangGraph CoAgent state is streamed as partial updates; error in action handler surfaces as a tool call error message in the chat thread.

Documentation covers React quickstart, useCopilotAction guide, context injection with useCopilotReadable, CoAgent (LangGraph) integration, self-hosted runtime setup, and Vercel/Next.js deployment guide. Prebuilt UI components reduce boilerplate. CopilotKit Cloud for managed inference. Confidence is docs-derived.