Leon

Overview:

Leon is an open-source personal AI assistant built around tools, context, memory, and agentic execution. It is designed to operate locally, use dedicated tools instead of relying on free-form guessing, and complete tasks from start to finish across deterministic workflows and agent-style execution. It is intended for developers and self-hosters who want a practical, privacy-aware assistant grounded in their real environment.

Core Features:

• Execution Modes: Supports smart mode (chooses how to handle a task), workflow mode (follows a fixed path), and agent mode (plans step by step).

• Tool-Based Execution: Uses real tools to get work done instead of only replying with plain text.

• Context Awareness: Uses context about the user's environment to keep answers grounded in what is happening on the machine and setup.

• Layered Memory: Maintains durable preferences, day-to-day context, and recent discussion context.

• Local and Remote AI Provider Support: Can work with both local models and remote AI providers to balance privacy, control, and capability.

• Modular Architecture: Organized as Skills -> Actions -> Tools -> Functions -> Binaries, with support for skills and toolkits covering search, productivity, system utilities, media, coding, and voice/audio features.

Use Cases:

• Running a privacy-aware assistant locally: Users can operate Leon with local models and local context instead of forcing everything through third-party services.

• Automating tasks with deterministic workflows: Using workflow mode to follow a fixed path for tasks that require step-by-step execution.

• Exploring agent-style task planning: Using agent mode to plan and execute tasks autonomously, with the ability to recover from errors.

• Extending the assistant with custom tools: Developers can build on top of the modular architecture using skills, toolkits, bridges, and binaries.

Why It Matters:

Leon is built on a modular architecture that prioritizes grounded behavior through explicit tools, context, and memory rather than vague model-only responses. It supports both local and remote AI providers, offering flexibility without committing to a single execution style. As an open-source project, anyone can inspect the architecture, build on top of it, and influence its direction, making it a transparent foundation for developing a personal AI assistant.