Private AI
The AI you can tell anything.
Last updated:
Because there's no one on the other end.
A cloud AI conversation starts by sending your words to a server, where the provider handles them under its own retention, review, and training settings. Loci doesn't take that route. The model runs on your device, so the prompt and the reply never need to reach anyone's service.
Incognito hides your history, not your question
A private browser window only hides where you've been from your own device. The question itself still goes out — through your internet provider, across the internet, to a server that receives exactly what you typed.
A local Loci chat never makes that trip. No provider receives it, and no account is attached to it. The safest data is the data that never moves.
What runs on the device
- Chat: understood and answered locally.
- Voice: processed on the device, not streamed to a server.
- Photos: image understanding happens on-device.
- No account required for any of it.
The online features are separate
Loci uses a connection to download model files and can perform optional web search when it is enabled and current information is needed. Search sends only the specific query directly from your device to the search services used for that platform and setting, not the whole conversation, and the request does not pass through a Loci server. On Windows, Loci tries Bing first, then DuckDuckGo and Wikipedia only as ordered fallbacks when an earlier provider has no usable results, and stops at the first provider with usable results. Windows can request up to four returned public HTTPS pages and use readable text from at most two. This does not mean every query contacts every provider. Optional anonymous diagnostics cover technical reliability and performance, never prompt or reply content, and can be turned off.
| What matters | Loci | Cloud AI |
|---|---|---|
| Where chat is processed | On your device. Conversations don't leave it. | Sent to remote servers to be answered. |
| Who's on the other end | No one. The model answers locally. | A provider receives the prompt and handles it under its own retention and training settings. |
| Account | None required for local AI. | Usually tied to an account and an identity. |
Using private AI on a Mac?
The private AI for Mac guide traces each data flow on its own, and shows how to prove it by pulling the Mac offline.
Privacy questions
Does Loci send my chats to a server?
No. Local conversations are processed on your own device, so chat data isn't uploaded to a Loci server.
Can anyone read my conversations?
For local chat, there's no server receiving them, so there's nothing on the other end to read.
Do I need an account?
No account is required for local AI.
Is any feature online?
Model downloads and optional web search use the internet. Search can run when it is enabled and Loci needs current information. Chat inference stays local.
For sensitive questions
Keep the question on your device.
Use Loci for the things you'd rather not hand to a cloud account.