Short answer: is Backyard AI private? In its local mode, yes, genuinely, structurally private, in a way almost no companion app can claim. That’s a refreshing thing to be able to write, because most “is this app private” questions end in “not really.” Backyard is the exception, and it’s worth understanding exactly why, plus the one caveat that changes the answer if you opt into it.
Why local mode is actually private
Backyard AI (formerly Faraday) is a desktop app that runs open-weight models on your own computer. When you chat in local mode, here’s what physically happens: the model file sits on your disk, your message is processed by your own CPU and GPU, and the reply comes back, without ever leaving the machine. There’s no server in the loop to receive your words, so there’s nothing to log, retain, or breach remotely.
This is the gold standard, and it’s the same property that makes any local setup trustworthy. We explain the underlying principle, privacy as an architectural property, not a policy promise, in our AI companion privacy guide. Backyard satisfies it because the computation happens on your hardware.
The one caveat: the optional cloud tier
Here’s the honesty the question deserves. Backyard, like several local apps, offers an optional cloud capability, a way to run larger models on remote hardware when your own machine can’t, sometimes called a tether or cloud tier. The moment you opt into that, the privacy math changes: your messages now travel to a server to be processed, exactly like any hosted app, governed by whatever that service’s policy says.
This isn’t a gotcha, it’s a feature for people whose hardware can’t run the big models, and it’s clearly a choice. But it means the accurate answer is conditional:
- Local mode: private by architecture. Your chats stay on your disk.
- Cloud/tether mode: private by policy, like any hosted service. Read the terms.
If maximum privacy is the goal, stay in local mode. If you turn on the cloud option, judge it the way you’d judge any cloud companion, on what’s committed in writing, not on the app’s local reputation.
How to verify the private claim yourself
You don’t have to take anyone’s word for it. The definitive test for any “local” app is the airplane-mode test:
- Make sure you’re running a local model (not the cloud tier).
- Disconnect from Wi-Fi / pull the network cable.
- Send a message.
If the companion still replies, the computation is happening on your machine and nothing is being uploaded, because nothing can be. A cloud app goes silent the instant you disconnect; a true local one keeps talking. We walk through this in detail in the offline AI girlfriend field guide.
Backyard’s privacy vs the cloud-companion default
| Typical cloud companion | Backyard (local mode) | |
|---|---|---|
| Where chats are processed | Their servers | Your machine |
| Stored where | Their database | Your disk |
| Breach exposure | Your chats + identity | Nothing to leak remotely |
| Trust model | Policy you can’t audit | Architecture you can verify |
By this comparison, Backyard in local mode is doing the right thing. The contrast isn’t Backyard versus some other local app; it’s any local app vs the cloud companions that retain everything by default, the category whose failure modes we map in are AI girlfriend apps safe.
The bottom line
Is Backyard AI private? In local mode, yes, and you can prove it with the airplane-mode test. The only asterisk is the optional cloud tier, which trades that architectural privacy for the convenience of bigger remote models; if you enable it, you’re back in policy-trust territory. And if what you liked about Backyard was that it ran on your machine, Ember keeps exactly that property and drops the assembly: local execution, no email, no account, and premium paid once in crypto rather than monthly. Either way, the headline is the good news: running it on your own machine is the private way, and Backyard’s local mode does exactly that.