Leon is an open-source personal AI assistant that lives on your own machine and gets things done through real tools rather than only answering questions. Started in 2019 as a classic voice assistant, it is being rebuilt as a modern agentic system with tools, context, layered memory and multiple execution modes. The project has earned about 17,600 stars on GitHub, and its 2.0 Developer Preview is the current focus of development.
Leon can run in three modes: smart mode, which picks an approach per request; controlled mode, which follows deterministic native skills and actions; and agent mode, which plans multi-step work. Its skills system spans search, productivity, system utilities, media and coding assistance, and a progressively loaded computer-use capability can operate desktop and browser interfaces and verify the visual result. Leon grounds its behavior in context about your actual environment, keeps layered memory for durable preferences and recent activity, and can work with local models or remote providers, balancing privacy against capability. It runs locally with a web app, and the project is candid that the 2.0 documentation is still catching up with the new core.
Leon is completely free and open source under the MIT license. Development is supported by sponsors; you only pay for any remote model providers you choose to connect.
Leon is for privacy-conscious users and tinkerers who want a self-hosted personal assistant they fully control, and for developers interested in contributing to an open, agentic assistant architecture.