Elissa Brings a Voice-Controlled AI Assistant to the Desktop

Summary

Elissa is a desktop voice assistant with an on-screen character for Windows, with macOS and Linux previews. Users can say “Hey Elissa” or press a keyboard shortcut to speak in English or Spanish, and the wake phrase is detected offline. The assistant supports OpenAI, Claude, Gemini, OpenRouter, Groq, Ollama, and LM Studio, so it can use cloud services or local models. It can read and edit PDFs, Word files, spreadsheets, CSV files, and code; operate applications and other windows; manage reminders and tasks; handle email; connect to MCP servers; control Spotify or other supported music players; and help inspect, edit, and test code. Every consequential action is shown in an approval card, including the exact file diff or command, and requires a real click. Voice input, the keyboard shortcut, and the selected AI model cannot approve an action. Elissa also refuses requests such as deleting files or email, reading password fields, opening programs or scripts, and commands that delete or discard work. The project stores conversations as Markdown memory notes and reports no account, telemetry, or proprietary server. With Ollama and on-device speech recognition, it says nothing leaves the computer. Windows 10/11 installers are available without administrator rights; macOS and Linux builds are previews. The project is MIT-licensed, and signed Windows distribution is still in progress.