Overview
Leon is an open-source personal AI assistant that goes beyond plain chat. Instead of only replying with text, it uses real tools, keeps memory of useful information, and can plan and execute tasks from start to finish.
Leon first launched in 2019 as a classic intent-classification assistant and is now being rebuilt into a more capable, agent-style system. It runs locally by default, can work with local models for privacy, and supports both local and remote AI providers so you can balance control and capability.
The project is currently focused on its 2.0 Developer Preview on the develop branch. A more stable, pre-agentic version still lives on the master branch for people who want the legacy experience.
What it does
- Three run modes: smart mode picks the approach for you, controlled mode follows deterministic native skills, and agent mode plans step by step
- Native skills for controlled actions plus agent skills backed by SKILL.md workflows
- Real tools so Leon can get work done instead of guessing in free-form text
- Layered memory that holds durable preferences, day-to-day context, and recent discussion
- Environment context so answers stay grounded in what is actually happening on your machine
- Support for both local and remote AI providers to balance privacy, control, and capability
Getting started
Leon needs Node.js 24 or newer and runs on Linux, macOS, and Windows. You clone the repo, install dependencies with pnpm, then start it locally.
Clone the repository
Clone Leon from GitHub and move into the project root.
git clone https://github.com/leon-ai/leon.git
cd leonInstall dependencies
Install pnpm globally, then install the project dependencies.
npm install --global pnpm@latest
pnpm installRun Leon
Start Leon. By default it runs locally and the app is available at http://localhost:5366.
pnpm startCheck your setup
Run the check command to confirm the setup went well.
pnpm run checkCommands and code are distilled from the project's own documentation — always check the official repo for the latest.
When to use it
- Running a private, self-hosted personal assistant that can use local models and local context instead of sending everything to third-party services
- Completing multi-step tasks through agent-style planning, native skills, and real tools rather than text-only answers
- Building and extending a modular assistant with custom skills, toolkits, and bridges across search, productivity, system utilities, coding, and voice
How Leon compares
Leon alongside other open-source assistants & chatbots tools AI/TLDR tracks, ranked by GitHub stars.
| Tool | Stars | What it does |
|---|---|---|
| OpenClaw | ★ 390k | OpenClaw is a self-hosted personal AI assistant that answers you on WhatsApp, Telegram, Slack, Discord, and many other channels, with voice and a live visual canvas. |
| Hermes Agent | ★ 246k | A self-improving personal AI agent from Nous Research that builds skills from experience, remembers across sessions, and reaches you on Telegram, Discord, Slack, and more. |
| Odysseus | ★ 87.3k | A self-hosted AI workspace that puts chat, agents, deep research, documents, email, notes, tasks and calendar behind one Docker Compose stack, over local or API models. |
| CowAgent | ★ 47k | A self-hosted assistant that plans and executes tasks with built-in file, terminal, browser and search tools, and answers across a web console plus a dozen messaging platforms. |
| AstrBot | ★ 40.6k | An all-in-one agent chatbot platform that puts LLM conversations, tools, knowledge bases and a plugin marketplace inside messaging apps like Telegram, Slack, Discord, QQ and Feishu. |
| OpenHuman | ★ 39.8k | A local-first desktop personal AI for macOS, Windows and Linux that keeps a compressed memory tree on your machine and orchestrates checkpointed research and automation workflows. |
| MindsHub | ★ 39.7k | An agent workspace for knowledge work and software development that runs swappable open-source agent harnesses against your choice of frontier or open models. |
| Leon | ★ 17.5k | Your open-source personal AI assistant, grounded in tools, context, and memory |
