Salvatore Sanfilippo — the engineer behind Redis, one of the databases quietly powering half the internet — just released something personal and precise. It's called ds4, and it's a tiny piece of software that does one thing: runs the DeepSeek V4 Flash AI model on a MacBook with 128GB of RAM. That's it. No monthly bill. No API. No data sent anywhere.
What makes this different from other "run AI locally" projects is how thoughtfully it's built. Most tools try to work with any AI model. This one was made specifically for one model, and that focus shows — it runs fast and handles extremely long conversations without breaking a sweat. It also uses your Mac's internal storage as a kind of overflow tank for memory, which means it can hold a truly enormous amount of context in a single session.
If you work with sensitive information — client contracts, financial records, internal strategy — running AI locally means that material never touches anyone else's servers. Not OpenAI's. Not Anthropic's. Nobody's.
This is still a tool for technical teams, but the fact that someone of Sanfilippo's calibre built it signals something: local AI is maturing fast, and the gap between cloud AI and your-own-hardware AI is closing.
Local inference — Running an AI model on your own computer, rather than sending your text to a company's server and getting a response back. Your data stays home.
Parameters — Roughly speaking, the "size" of an AI model's brain. More parameters often means more capable reasoning. 284 billion is frontier territory.
Context window — How much of a conversation the AI can hold in mind at once. A 1-million-token context window means you could feed it an entire book and it wouldn't forget the beginning by the time it reaches the end.
KV cache — The AI's short-term working memory during a conversation. Usually it lives in RAM; this tool spills it onto disk so you don't need an obscene amount of memory.
Something to sit with: if AI this capable can run on hardware you already own, what does that change about which conversations you'd be willing to have with it?