leon: an open-source personal AI assistant with agentic execution and local-first privacy

What it solves

Leon is an open-source personal AI assistant designed to be practical, privacy-aware, and grounded in the user's real environment. It moves beyond simple intent-classification to handle complex goals by using tools, memory, and agentic execution to complete tasks from start to finish.

How it works

Leon operates using a modular architecture consisting of a main runtime server, a web application, and a system of bridges (Node.js and Python) that connect to tools and binaries. It supports three execution modes:

• Smart mode: Automatically chooses the best execution path.
• Controlled mode: Follows deterministic native skills and actions.
• Agent mode: Plans and executes tasks step-by-step.

To stay grounded, Leon uses layered memory (durable preferences, daily context, and recent discussions) and a proactive pulse system. It can be powered by either local or remote AI providers to balance privacy and capability.

Who it’s for

It is intended for users who want a private, extensible personal assistant that can interact with their local machine and environment rather than relying solely on cloud-based LLMs.

Highlights

• Agentic Execution: Ability to plan and execute multi-step tasks using tools.
• Privacy-First: Supports local AI models and local context to avoid third-party dependency.
• Modular Extensibility: Uses a system of skills, toolkits, and bridges to add new capabilities.
• Hybrid Execution: Combines deterministic workflows with flexible agent-style planning.