cherry-studio

Cherry Studio abstracts 50+ LLM providers (OpenAI, Anthropic, DeepSeek, Ollama, etc.) through a unified API service layer that handles provider-specific parameter construction, API key rotation, and streaming response normalization. The Provider System maps model configurations to provider-specific implementations, enabling seamless switching between providers without changing application logic. This is implemented via a service-oriented architecture where each provider has a dedicated adapter that translates Cherry Studio's canonical request format into provider-specific API calls.

Cherry Studio implements an Agent System that orchestrates multi-step reasoning workflows by decomposing user intents into subtasks, executing tools via the Model Context Protocol (MCP), and managing agent state across iterations. Agents can invoke MCP tools (code execution, file operations, web search) through a standardized tool registry, with responses fed back into the reasoning loop. The MCP Architecture manages server lifecycle, tool discovery, and execution sandboxing, while the Agent System maintains conversation context and decision history across multiple reasoning steps.

Cherry Studio exposes a local API server that enables external applications to interact with the application via HTTP. The Local API Server provides REST endpoints for chat, assistant management, and knowledge base operations. OAuth Integration enables secure authentication for API access, supporting both local and cloud-based OAuth providers. LAN Transfer and File Management enables users to transfer files between devices on the same network without cloud storage, using local network discovery and peer-to-peer transfer.

Cherry Studio includes a Notes and Rich Text Editor that enables users to create and edit rich text documents with markdown support. The editor supports inline formatting (bold, italic, code), lists, tables, and code blocks with syntax highlighting. Notes are persisted to the local database and can be linked to conversations or assistants. The system provides a WYSIWYG editing experience with markdown preview, enabling users to write documentation or notes alongside AI conversations.

Cherry Studio implements a Theme and Localization system that supports multiple languages (English, Chinese, etc.) and theme modes (light, dark, auto). The system uses a localization framework to manage translated strings, with language selection persisted in settings. Theme switching is implemented via CSS variables and React context, enabling instant theme changes without page reload. The system respects system theme preferences and enables manual override.

Cherry Studio implements an Auto-Update System that checks for new versions in the background, downloads updates, and prompts users to install. The system uses electron-updater for update management, with support for staged rollouts and update channels (stable, beta). Updates are downloaded in the background without blocking the application, and users can defer installation until a convenient time. The system maintains version history and enables rollback to previous versions.

Cherry Studio implements a Selection Assistant that integrates with the system context menu, enabling users to select text anywhere on the system and send it to Cherry Studio for analysis or processing. The system uses Electron's native context menu APIs to register custom menu items. When text is selected, users can choose from predefined actions (translate, summarize, explain, etc.) which are executed by the appropriate assistant. Results can be displayed in a floating window or copied to clipboard.

Cherry Studio integrates image generation capabilities through connected LLM providers that support image generation (DALL-E, Midjourney, etc.). The Paintings and Image Generation system enables users to generate images from text prompts within the chat interface. Generated images are displayed inline in conversations and can be saved or edited. The system supports image-to-image editing and variation generation. Integration with MCP tools enables advanced image processing (upscaling, background removal, etc.).

Cherry Studio implements a Knowledge Base System that ingests documents (PDFs, markdown, text) through a document processing pipeline with OCR support, generates embeddings for semantic search, and retrieves relevant context for RAG-augmented chat. The system stores documents in a local database with metadata indexing, enabling full-text and semantic search across ingested knowledge. When a user queries, the system retrieves relevant document chunks and injects them into the LLM context, enabling the model to answer questions grounded in custom knowledge.

Cherry Studio provides an Assistant System that enables creation of custom AI assistants with predefined roles, system prompts, and skill compositions. Assistants are defined as configuration objects specifying model, provider, system prompt, and enabled skills. The system includes 300+ pre-built assistants covering various domains (coding, writing, analysis, etc.). When a user selects an assistant, Cherry Studio injects the assistant's system prompt and skills into the conversation context, enabling role-specific behavior without requiring manual prompt engineering.

Cherry Studio implements a Topic and Session Management system that organizes conversations into topics (folders), with each topic containing multiple chat sessions. Conversations are persisted to a local database with full message history, metadata (timestamps, model used, tokens), and session state. The Message System handles message rendering, streaming responses, and markdown/code block formatting. Redux state management maintains the current session context, enabling seamless switching between conversations while preserving full history and context.

Cherry Studio implements a Streaming and Message Processing system that handles real-time LLM responses through chunked streaming, with progressive message rendering and live token counting. The system normalizes streaming responses across different providers (OpenAI, Anthropic, etc.) into a unified stream format, enabling consistent UI updates. As tokens arrive, the system updates the UI incrementally, displays token usage in real-time, and handles stream interruption/cancellation. Markdown and code blocks are parsed and rendered progressively, enabling users to see formatted output as it streams.

Cherry Studio implements a comprehensive MCP Architecture that manages MCP server lifecycle (startup, shutdown, health checks), discovers available tools and resources from connected servers, and executes tools with request/response handling. The MCP Server Management system maintains connections to multiple MCP servers, handles server configuration (local or remote), and manages authentication. MCP Tools and Execution provides a unified interface for invoking tools across different servers, with error handling and timeout management. MCP Prompts and Resources enable servers to inject structured context (prompts, file resources) into agent reasoning.

Cherry Studio integrates code execution capabilities through OpenClaw (a code execution sandbox) and provides syntax-aware code analysis. The Code Tools system enables agents to execute code snippets in isolated environments, with output capture and error handling. Code blocks in chat are rendered with syntax highlighting for 40+ languages, with copy-to-clipboard and execution buttons. The system maintains code execution history and enables debugging through output inspection. Integration with MCP tools enables code execution across multiple languages (Python, JavaScript, Shell, etc.).

Cherry Studio integrates web search capabilities that enable agents and assistants to retrieve real-time information from the internet. The Web Search Settings configure search provider (Google, Bing, etc.) and enable/disable search per assistant. When web search is enabled, agents can invoke search tools to find current information, with results including source URLs and snippets. Search results are injected into the LLM context, enabling responses grounded in current web information. The system handles search result ranking and deduplication.

Cherry Studio is built on Electron's multi-process architecture, separating the main process (privileged operations) from renderer processes (UI). The IPC Communication System provides bidirectional message passing between main and renderer processes using a channel-based architecture defined in IpcChannel.ts. State Management with Redux maintains application state in the renderer process, with persistence to the local database via main process IPC calls. The architecture enables responsive UI by offloading heavy operations (file I/O, API calls, database queries) to the main process.