assistant-ui

Provides a system of unstyled, composable React components (Thread, Message, Composer, ActionBar) built on Radix UI primitives that can be assembled into custom chat interfaces without enforcing a specific visual design. Uses a context-based state management pattern where each component subscribes to a centralized store, enabling fine-grained control over rendering and behavior while maintaining separation of concerns between logic and presentation layers.

Implements a streaming infrastructure (@assistant-ui/react-data-stream) that handles real-time message chunks from AI backends using a protocol-agnostic message format. Uses message accumulation with configurable throttling to batch incoming chunks, preventing excessive re-renders while maintaining perceived responsiveness. Supports both text streaming and structured tool call streaming with automatic conversion between different message formats (OpenAI, Anthropic, LangGraph).

Provides React Native bindings (@assistant-ui/react-native) that enable building chat UIs for iOS and Android using the same component API as web. Uses React Native's native components (ScrollView, TextInput, etc.) under the hood while maintaining API compatibility with web components. Supports streaming, tool execution, and state management on mobile platforms with platform-specific optimizations for performance and battery life.

Provides React Ink bindings (@assistant-ui/react-ink) that enable building chat UIs for terminal/CLI applications using the same component API as web and mobile. Uses React Ink's terminal rendering engine to display messages, composer input, and action bars in the terminal. Supports streaming, tool execution, and keyboard navigation optimized for terminal environments.

Provides a CLI tool (@assistant-ui/cli) for scaffolding new chat projects, installing components, and running codemods for migrations. Uses AST-based transformations to automatically update code when upgrading between versions, handling breaking changes without manual refactoring. Supports interactive component installation with customization options and project template generation.

Provides pluggable content rendering system with built-in support for markdown (@assistant-ui/react-markdown) and code syntax highlighting (@assistant-ui/react-syntax-highlighter). Uses a renderer registry pattern where different content types (text, markdown, code, custom) can have custom rendering implementations. Supports streaming markdown rendering (progressive rendering as markdown arrives) and automatic language detection for code blocks.

Provides development tools (@assistant-ui/react-devtools) for debugging chat state, message flow, and component rendering. Includes an MCP (Model Context Protocol) documentation server that exposes assistant-ui's API and component documentation for AI-assisted development. DevTools UI shows real-time state updates, message history, and performance metrics. MCP server enables AI tools to query documentation and generate code.

Provides Python packages for building assistant-ui backends, including message format conversion, streaming utilities, and integration with Python AI frameworks (LangChain, LangGraph). Enables building chat backends in Python while using assistant-ui for the frontend, with automatic format conversion between Python and JavaScript representations. Supports streaming responses and tool execution from Python backends.

Provides a runtime system for executing tool calls with built-in support for human approval workflows. Uses a tool registry pattern where tools are defined with schemas, execution handlers, and optional approval requirements. Supports streaming tool parameters (for tools that accept large inputs), automatic error handling with retry logic, and state management for tool execution status (pending, executing, completed, error). Integrates with the message streaming system to handle tool calls that arrive mid-stream.

Abstracts the runtime layer to support multiple AI backends (Vercel AI SDK, LangGraph, custom implementations) through a unified interface. Uses adapter patterns to convert between different message formats (OpenAI, Anthropic, LangGraph) and provides runtime implementations that handle backend-specific details like streaming protocols, tool call formats, and error handling. The @assistant-ui/react-ai-sdk and @assistant-ui/react-langgraph packages provide pre-built adapters for common backends.

Provides a pluggable attachment system for handling file uploads, previews, and processing. Uses an adapter pattern where different file types (images, documents, code) can have custom handlers for preview rendering and content extraction. Supports both client-side and server-side processing, with automatic MIME type detection and validation. Attachments are integrated into the message stream and can be referenced by tool calls.

Implements a centralized state management system (@assistant-ui/store) using React Context and custom hooks, providing access to chat state (messages, thread, composer input) without prop drilling. Uses a subscription-based pattern where components subscribe to specific state slices, enabling fine-grained re-render optimization. Provides hooks like useThreadState, useMessageState, useComposerState for accessing and mutating state, with built-in support for undo/redo and message editing.

Provides a thread-based message storage system that organizes conversations into threads with support for branching, editing, and deletion. Uses an immutable message tree structure where each message has a unique ID and optional parent reference, enabling non-linear conversation flows (e.g., exploring alternative AI responses). Supports message editing with automatic history tracking and deletion with cascade behavior. Thread state is managed centrally and can be persisted to external storage.

Provides a registry of pre-built, styled chat components (Thread, Message, Composer, ActionBar) built on shadcn/ui and Radix UI primitives. Components are fully customizable through props and CSS, with automatic dark mode support and accessibility features (ARIA labels, keyboard navigation). The component registry system allows registering custom components to override defaults, enabling white-label implementations. Components are available as copy-paste code (shadcn-style) or npm packages.

Provides a pre-built adapter (@assistant-ui/react-ai-sdk) that integrates with Vercel AI SDK's useChat and useCompletion hooks, handling message streaming, tool calling, and provider abstraction. Automatically converts between Vercel AI SDK's message format and assistant-ui's internal format, enabling seamless use of Vercel's provider ecosystem (OpenAI, Anthropic, Cohere, etc.). Supports streaming tool calls and automatic tool execution through the HITL system.

Provides a pre-built adapter (@assistant-ui/react-langgraph) that integrates with LangGraph for complex, graph-based AI workflows. Handles streaming of graph execution events, tool call execution within the graph context, and automatic message extraction from graph state. Supports both synchronous and asynchronous graph execution with real-time status updates. Enables building multi-step agentic workflows with branching, loops, and conditional logic.