Control is underrated. You don't need a Ferrari to go buy milk. And you don't need a massive, data-hungry cloud AI for every simple request. The Leon AI philosophy: - Default: run locally on your hardware (privacy + speed). - Fallback: only call the big cloud APIs when the task

🤯 WOW! I am actually speechless. I just witnessed Leon AI self-evolve in real-time. The challenge: I asked Leon to write the code for a "Podcast Generator" skill. He wrote it. He installed it. Then... I asked him to use it. The result: "Leon, generate a podcast about how

🏗️ A note on "Progress." I see the DMs asking for updates after ~36 hours of quiet. Here is the reality of senior engineering: Shipping code is the final step, not the only step. - 80% of the work is architecture, planning, and vision alignment. - 20% of the work is typing the

What's on the menu. Zoom in for the alpha...

To avoid pricey surprises... When using third-party tools, the AI entity must provide an estimation cost at runtime. The owner must have the option to confirm the run. There should be a standard in the industry. That's one more TODO for Leon AI.

Facts. Approaching 17k GitHub stars. Leon AI isn't a "quick" project. And it's only the beginning. Wait until I hit the next few milestones and unlock new contributions from the community. You have been warned.

Dad mode: ON! (don't mind the beautifier filter from the Xiaomi phone)

When installing Leon AI, it will automatically download and install several NVIDIA (including a specific CUDA runtime) and PyTorch prebuilt libraries. I developed this so when Leon uses tools that leverage AI models, you won't even need any Python runtime or anything installed