What is Sia, and does my data stay private?
Sia is a personal AI model that installs and runs on your own computer. The model weights, your conversation history, and the data it learns from all live on your machine, so your data stays private by design. No shared cloud does the inference. The only things that leave your machine are integrations you explicitly turn on, like checking your calendar. Sia is in beta and open for pre-order, with delivery targeted for August 2026.
The single most common question about Sia is some version of “if it’s AI, where does my stuff go?” The short answer: nowhere you didn’t send it. Here’s the longer version, with the caveats included.
What is Sia?
Sia is a personal AI model that installs on your own computer. It’s the compact, distilled version of Physea’s architecture, small and fast enough to run on real consumer hardware today. Once it’s installed, it lives on your machine and keeps adapting to your work the more you use it.
That last part is the actual point. Most AI you’ve used was trained once and frozen. Sia changes the model itself as it learns what you correct, what you accept, and what you ask about. Day one it’s generally capable. Six months in, it writes more like you, knows your project history, and stops over-explaining things you already know.
Is my data private with Sia?
Yes, and the reason matters: with Sia, privacy is structural rather than a policy you have to trust. Here’s what that means in practice.
| What | Where it lives |
|---|---|
| The model weights | On your machine |
| Your conversation history | On your machine |
| The data it learns from | On your machine |
| The adaptation log (what it changed) | On your machine |
There’s no shared inference layer, no central server doing the thinking and seeing your prompts along the way. With a hosted assistant, every message you send travels to someone else’s computer to be answered. With Sia, the answering happens locally.
So what does leave my machine?
Being honest here is the whole job. Two things can send data out, and both are under your control:
- Integrations you connect. If you wire up your calendar, email, or Drive, the model can call those apps directly over standard OAuth. That traffic goes where it has to go to do the task. Each connection is scoped per app and revocable any time.
- Nothing else. And if you want to be certain, there’s an offline toggle that seals every external call. Flip it and the model can’t reach out at all.
So the accurate statement isn’t “no data ever leaves.” It’s “nothing leaves unless you explicitly connect it, and you can shut even that off.”
Isn’t this just ChatGPT’s memory feature?
No, and the difference is real. ChatGPT memory is closer to a sticky note: it writes a short summary and pastes it into your next prompt, but the underlying model stays exactly the same. Sia changes the model. One is reading a note off the fridge; the other is actually learning.
If you want the hardware side of this (what size machine runs what) see what AI model can my computer run.
When can I get it, and what does it cost?
Sia is in beta, with delivery targeted for August 2026. It’s a one-time purchase rather than a subscription, and it’s the only Physea model currently open for pre-order. When you reserve it, your card is saved but not charged until your install is built and ready. For the full breakdown of why there’s no monthly fee, read how Physea’s pricing works.
Questions the FAQ didn’t answer? Email [email protected].
Common questions
- Does Sia send my data to the cloud?
- No. The model runs locally and your data stays on your machine. The only outbound traffic is from integrations you choose to connect, like a calendar over OAuth. There's an offline toggle that seals every external call.
- When does Sia ship?
- Sia is in beta with delivery targeted for August 2026. You can pre-order now; your card isn't charged until your install is built and ready.
- How is Sia different from ChatGPT or Claude?
- Hosted models like ChatGPT and Claude run on someone else's servers and don't change for you. Sia runs on your hardware and actually adapts the model to your work over time. Nothing you type leaves your machine unless you connect an integration.
- Is Sia just a local version of an open model like Llama?
- No. Tools like Ollama run generic open models that don't know you. Sia is tuned to your work and keeps reshaping itself on your machine as you use it.