Your MacBook Can Now Run a Powerful AI Chatbot, 3x Faster

A company called Inco has released a free piece of software called Splash that lets a modern MacBook run a full AI assistant by itself — no internet, no accounts, no monthly bill. The assistant is a 27-billion-parameter model called Qwen3.8, which is roughly the size of a mid-range chatbot. Everything you type stays on your laptop.

Speed is the headline here. Splash hits 144 tokens per second, which is faster than most people can read. For comparison, the popular tool Ollama runs about three times slower, and Inco says Splash is nearly four times faster when an "agent" (AI that can take actions on your behalf, not just chat) splits work across several helper versions of itself. In practice, that's the difference between watching a chatbot slowly type and having an actual back-and-forth conversation.

The catch is hardware. Splash needs an M3-or-newer Mac, the latest macOS, and at least 36GB of memory — essentially a recent high-end MacBook Pro or a well-loaded desktop Mac, which typically starts around $2,000. If you have one, setup is a single command, and there's also a friendly app version inside LM Studio for people who don't like typing commands. It works with popular coding assistants like Claude Code and Codex.

The bigger picture: running AI locally used to be a hobbyist project that produced slow, dumb results. This shows it's becoming genuinely usable. You get privacy — nothing you type leaves your machine — and you never pay a subscription. What you don't get is the raw brainpower of the biggest cloud models, which still handle the hardest questions better. But the gap is narrowing, and the trade-off is looking better every few months.