Versions:
Witsy, published by Nicolas Bonamy and currently at version 3.5.2 across a release history of 35 versions, is a desktop AI assistant and universal MCP client. The application follows a BYOK (Bring Your Own Keys) model, requiring users to supply API keys for the LLM providers they wish to use, though Ollama integration allows models to run locally for free. Witsy positions itself as one of the first universal MCP clients, enabling Anthropic MCP servers to run with virtually any LLM. Supported providers include OpenAI, Ollama, Anthropic, MistralAI, Google, xAI, Azure, OpenRouter, DeepSeek, Groq, and Cerebras, with additional providers such as Together, SiliconFlow, and Fireworks connectable through an OpenAI compatibility layer. Core capabilities cover chat completion with vision model support for image description, text-to-image and text-to-video generation via OpenAI, Google, xAI, Replicate, fal.ai, and HuggingFace, plus image-to-image editing and image-to-video conversion through Google, Replicate, and fal.ai. LLM plugins extend functionality with features such as Python code execution and Internet search, while a long-term memory plugin increases the relevance of responses. Productivity-oriented tools include a Scratchpad for interactively refining content with any model, Prompt Anywhere for generating content directly within other applications, AI commands that operate on highlighted text in almost any application, and Experts prompts that specialize the assistant on a given topic. The software also supports chatting with local files and documents through RAG, transcription and dictation via speech-to-text, realtime chat, and read-aloud playback of assistant messages or text from other applications, with the latter requiring an OpenAI or ElevenLabs API key. Witsy fits the AI assistant and productivity software categories, serving users who want a single desktop client for multi-provider LLM access, content generation, document interaction, and MCP-based tooling.
Tags: