Wisp AI Assistant

Wisp is an open-source, hotkey-driven AI overlay for the desktop, written entirely in Python and released under the MIT license. Instead of switching to a chat app, you highlight text, press a hotkey, pick an action, and Wisp streams the answer into a compact bubble or pastes it back at your cursor. It can capture exactly the context you allow: selected text, clipboard, the focused app, open documents, browser pages, recent files, or a drawn screen snip sent to a vision model. Voice is built in, with local speech-to-text via faster-whisper and on-device or cloud TTS. Wisp is local-first: there is no hosted storage layer, provider keys live in your OS keychain, optional memory is stored locally with a viewer, and privacy mode can warn or redact before sensitive context leaves your machine. You bring your own model provider — Anthropic, OpenAI, Google, Groq, OpenRouter, Mistral, local Ollama or any OpenAI-compatible server, and many more. MCP support is prominent and two-way: a bundled MCP bridge addon acts as an MCP client, exposing any Model Context Protocol server's tools to Wisp's model, while the Wisp Context Server is a local MCP stdio server that lets trusted clients like Claude Desktop, Cursor, and Codex read live desktop context through five read-only tools. A sandboxed addon system and an agent task framework round out an unusually extensible free assistant for Windows 10+, macOS 13+, and Linux (X11, with Wayland in progress).