leon

Leon is an open-source personal AI assistant designed to run locally, use explicit tools, retain context, and complete tasks through deterministic workflows or agent-style planning. As of the repository notice, the `develop` branch is focused on the Leon 2.0 Developer Preview. Its new documentation is not ready, and the public documentation site largely describes the legacy architecture. The `master` branch remains available for the older, more stable pre-agentic release. The current assistant supports three execution modes: `smart`, which selects an approach automatically; `controlled`, which follows deterministic native skills and actions; and `agent`, which can plan tasks step by step. It includes native skills and `SKILL.md`-backed agent skills, with a native skill flow described as Skills → Actions → Tools → Functions → Binaries. Leon is built around environment context, layered memory for durable preferences and recent interactions, a compact self-model, and a bounded proactive pulse system. It can use local or remote AI providers and includes progressively loaded computer-use capabilities for desktop and browser interfaces, with visual verification and support for running locally or through Satellite. The repository is organized into a main server runtime and HTTP API, a web application, UI components, native and agent skills, Node.js/Python bridges, Python TCP services, and generated architecture context documents. Included skills and toolkits cover search, productivity, system utilities, media workflows, coding assistance, memory-backed interactions, and voice/audio features. Requirements listed for setup are Node.js 24 or newer, pnpm, and Linux, macOS, or Windows. After installing dependencies with `pnpm install`, Leon can be started with `pnpm start` and is exposed by default at `http://localhost:5366`; `pnpm run check` verifies the setup. The project is MIT licensed and accepts community involvement through its roadmap, Discord, issues, newsletter, and contributor onboarding process.