What is this?
Leon is an open-source personal AI assistant that runs locally and is built around tools, context, memory, and agentic execution — it understands a goal, uses real tools, and completes tasks end to end.
Leon is an open-source personal AI assistant that runs locally and is built around tools, context, memory, and agentic execution — it understands a goal, uses real tools, and completes tasks end to end.Is this a web app, a desktop program, or something I host myself?It is a self-hosted stack: a Node.js server (runtime, memory, HTTP API) plus a web application you open in a browser, with some Python services in the runtime.
It is a self-hosted stack: a Node.js server (runtime, memory, HTTP API) plus a web application you open in a browser, with some Python services in the runtime.How do I get it running?Install Node.js 24 or newer (Volta is recommended), clone the repo, and run Leon — by default everything runs locally and you use it through the web app.
Install Node.js 24 or newer (Volta is recommended), clone the repo, and run Leon — by default everything runs locally and you use it through the web app.Do I need anything besides Node.js?Node.js >= 24 is the core requirement, but part of the stack is Python (bridges and some services). You also pick an AI provider — local or remote; OpenAI, Anthropic, Groq, Cerebras and others are wired in.
Node.js >= 24 is the core requirement, but part of the stack is Python (bridges and some services). You also pick an AI provider — local or remote; OpenAI, Anthropic, Groq, Cerebras and others are wired in.Is the current version stable enough to rely on?Not yet — Leon is currently a 2.0 Developer Preview on the develop branch and the new docs are not ready. If you want the older, more stable pre-agentic version, use the master branch.
Not yet — Leon is currently a 2.0 Developer Preview on the develop branch and the new docs are not ready. If you want the older, more stable pre-agentic version, use the master branch.Does everything go through the cloud, or can it stay on my machine?Privacy is a stated design goal: Leon can work with local models and local context instead of forcing everything through third-party services, while remote providers stay an option.
Privacy is a stated design goal: Leon can work with local models and local context instead of forcing everything through third-party services, while remote providers stay an option.What can it actually do out of the box?The repo already ships skills and toolkits for search, productivity, system utilities, media workflows, coding assistance, memory-backed interactions, and voice/audio features. Native skills handle controlled actions, agent skills follow SKILL.md workflows.
The repo already ships skills and toolkits for search, productivity, system utilities, media workflows, coding assistance, memory-backed interactions, and voice/audio features. Native skills handle controlled actions, agent skills follow SKILL.md workflows.How does it decide what to do — is it just chat replies?Three modes: smart mode chooses for you, controlled mode follows deterministic native skills and actions, and agent mode can plan step by step. Leon uses real tools instead of only replying with plain text.
Three modes: smart mode chooses for you, controlled mode follows deterministic native skills and actions, and agent mode can plan step by step. Leon uses real tools instead of only replying with plain text.Can it control my computer and browser?Yes — it has progressively loaded computer use that can operate desktop and browser interfaces and verify visual outcomes, running locally or through Satellite without exposing its full tool surface to unrelated turns.
Yes — it has progressively loaded computer use that can operate desktop and browser interfaces and verify visual outcomes, running locally or through Satellite without exposing its full tool surface to unrelated turns.Why build an open-source assistant instead of using a big-tech one?The README lists four reasons: privacy (local models over third-party services), grounded behavior (tools, context and memory over vague model-only replies), extensibility (skills, toolkits, bridges, binaries) and open source.
The README lists four reasons: privacy (local models over third-party services), grounded behavior (tools, context and memory over vague model-only replies), extensibility (skills, toolkits, bridges, binaries) and open source.What are similar projects?Dodaios, fullstack-agent, openclaw, Tel-Agent, claude-code, OpenDots, eliza, brigade
Dodaios, fullstack-agent, openclaw, Tel-Agent, claude-code, OpenDots, eliza, brigadeWhich topics is it in?personal-assistant, assistant, skill-md, voice-assistant, speech-synthesis, voice, bot, computer-use, ai-assistant, speech-to-text, speech-recognition, text-to-speech
personal-assistant, assistant, skill-md, voice-assistant, speech-synthesis, voice, bot, computer-use, ai-assistant, speech-to-text, speech-recognition, text-to-speech