SuperAPI AI Directory
Partner with us中文AI newsFavorites
All toolsAI developer platformsWitsy

Witsy

Witsy is an AI desktop assistant supporting models from all major providers and one keyboard shortcut away!

Open website

About this tool

Witsy is a desktop AI assistant and also a universal MCP (Model Context Protocol) client. It supports models from all major providers and can be invoked at any time with a single keyboard shortcut.

Purpose

Witsy is designed to integrate AI capabilities into desktop workflows, allowing users to use multiple AI models for chat, content generation, image processing, speech-to-text, and other tasks without switching applications. It uses a BYOK (Bring Your Own Keys) model, where users need to provide their own API keys for LLM providers; models can also be run locally for free using Ollama.

Main capabilities

  • Chat completion: Supports vision models (describing images).
  • Image and video generation: Text-to-image, text-to-video, image-to-image (image editing), image-to-video.
  • LLM plugins: Execute Python code, search the internet, etc., to enhance LLM capabilities.
  • MCP support: Supports Anthropic MCP servers, and is one of the few universal MCP clients, allowing MCP servers to work with almost any LLM.
  • Scratchpad: Interactively create the best content.
  • Prompt Anywhere: Generate content directly in any application.
  • AI commands: Perform quick actions on selected text in almost any application.
  • Expert prompts: Keep bots focused on a specific topic.
  • Long-term memory plugin: Improve the relevance of LLM responses.
  • Read aloud: Read assistant messages or text in any application aloud.
  • RAG: Chat with local files and documents.
  • Speech-to-text/dictation: Supports multiple speech-to-text models.
  • Realtime chat/voice mode.
  • Anthropic Computer Use support.
  • Local conversation history: Automatically generates titles.
  • Code formatting and copy to clipboard.
  • Conversation PDF export.
  • Image copy and download.

How to use

  1. Download and install: Download the binary from the releases page, or on macOS use brew install --cask witsy.
  2. Configure API keys: To use OpenAI, Anthropic, Google, or Mistral AI models, enter an API key; to use Ollama models, install Ollama and download a model; to use text-to-speech, an OpenAI API key is required; to use internet search, a Tavily API key is required.
  3. Prompt Anywhere: In any editable content, press the shortcut (Shift+Control+Space / ^⇧Space), enter a prompt, and Witsy will type the text directly into the application. On Mac, experts can be triggered automatically based on the foreground application.
  4. AI commands: Select any text, press the shortcut (Alt+Control+Space / ⌃⌥Space), and choose a command to execute. You can also create custom commands.
  5. Chat with documents (RAG): Click the database icon on the left side of the prompt box, manage and create document repositories, and after adding documents, select the repository.
  6. Dictation/speech-to-text: Supports models such as GPT4o-Transcribe, Speechmatics, Groq Whisper V3, Fireworks.ai Realtime Transcription, fal.ai Wizper V3, fal.ai ElevenLabs, nVidia Microsoft Phi-4 Multimodal, and more. After transcription, you can copy, translate, or insert it into the previous application.
  7. HTTP API: The local HTTP API runs by default on port 8090 on localhost (if occupied, the next available port is used). Supports GET and POST requests.
  8. Command-line interface (CLI): Automatically installed on first launch of Witsy. On macOS/Linux, create a symbolic link /usr/local/bin/witsy, and on Windows, add it to the user PATH. Available commands include /help、/model、/port、/clear、/history、/exit. The CLI requires the Witsy desktop app to be running and the HTTP API server to be enabled (default port 4321).
  9. Agent Webhook: Create or select an agent in Agent Forge, check Webhook on the Invocation tab, generate a unique token, and use http://localhost:{port}/api/agent/run/{token} Triggered. Supports GET/POST, with parameters passed as input variables to the agent prompt. Returns runId and statusUrl for checking execution status.

Use Cases

  • Users who need to quickly invoke AI across multiple desktop applications.
  • Users who want to use their own API keys or local models (Ollama).
  • Developers who need to combine MCP servers with different LLMs.
  • Users who need speech-to-text, dictation, and real-time voice chat.
  • Users who need RAG conversations with local documents.
  • Users who need to integrate AI capabilities into automated workflows via HTTP API or CLI.

Limitations

  • Uses a BYOK model; using cloud models requires providing your own API key.
  • Some features (such as text-to-speech and internet search) require additional API keys.
  • The local Whisper model requires downloading large files.
  • The HTTP server listens only on localhost by default; external access requires a reverse proxy and authentication configuration.
  • The CLI requires the Witsy desktop application to be running.
  • Some features (such as automatically selecting an expert based on the foreground application) are Mac-only.
  • Detailed information such as specific charges, pricing, and platform support is not yet provided.

Related tools

Browse category