Witsy

Witsy is an open-source desktop application that serves as a universal AI assistant and MCP client. It operates on a bring your own key model, requiring users to provide API keys for various LLM providers or use Ollama to run models locally. The application is available as a desktop binary for multiple platforms and can be installed on macOS via Homebrew.

The software integrates a wide array of AI capabilities into a single interface, supporting chat, image and video generation, and speech processing. It allows users to connect different AI backends to a unified workspace, enabling the use of Model Context Protocol (MCP) servers across different LLMs. This versatility allows for a customized AI setup where the user controls the specific model and provider used for each task.

Key features

• Support for multiple providers including OpenAI, Anthropic, Google Gemini, and DeepSeek
• Universal MCP client capabilities to run MCP servers with any LLM
• Prompt Anywhere feature to generate content directly into other applications
• Local RAG support for chatting with local files and documents
• Integration with Ollama for local model execution
• AI commands for processing highlighted text in external applications
• Support for text to speech, speech to text, and vision models
• Local HTTP API for triggering commands from external applications

Built with Electron and Vue.js, Witsy is designed for users who want a centralized hub for various AI models without being locked into a single ecosystem. It includes a scratchpad for iterative content creation and a long term memory plugin to improve response relevance. The application also supports Anthropic Computer Use and provides a local history of conversations with automatic titling. For advanced users, it offers a local HTTP API running on port 8090, which allows external applications to trigger chat, transcription, or prompt functions via GET and POST requests.

Witsy is a flexible, provider agnostic interface for power users of generative AI.