Your intelligence, on your own device.
The whole chanio loop — pull your day's chats down from the LLMs you already use, store them on your hard drive, derive your passions, let the models work them overnight, wake to new insight — runs on Apple's on-device stack. Your chats happen in the cloud; the synthesis happens at home. Not in a cloud that “protects” your intelligence by holding it. On the iPhone and Mac you already own. About $0 a night. The memory it builds never leaves your device.
Everything you did with AI, working for you while you sleep.
Seven beats. Each one runs on your device. You only ever see the first and the last — you use your AI, and you wake up smarter.
The system, mapped to Apple's real primitives.
None of this is a metaphor. Each piece of the loop runs on a specific, shipping Apple API. The model is on your device; the overflow is cryptographic; the sync is your iCloud, not our server. Tap any of them for Apple's own documentation.
The model is rented. The memory is yours.
Apple itself rents a frontier model for Siri. Everyone runs the same commodity intelligence now — the value moved up, to the layer that remembers you and compounds. The cloud giants know it, so they all offer to “amplify and protect your intelligence.” Read the fine print: it lives inside their systems, and “protect” means a promise not to train on it.
chanio takes the opposite vow. Your intelligence never leaves your hard drive. There is nothing to protect it from, because we never hold it. Apple's on-device model does the work where your memory already lives — the agents come to you. No AI should eat the intelligence of the person it's deployed inside.
Running now on Mac. Shipping on iPhone with iOS 27.
This is not a mockup. The full loop — capture, passion-derivation, the on-device memory pack, a local model answering from it — runs today on the founder's Mac, at $0 for everything but the one daily frontier call. And as of this week the same pack runs on Apple's own on-device model through the runtime Apple now ships in macOS 27 — fm respond, the memory pack loaded as its instructions. In about five seconds, on-device, it answered a grounded question about the person — citing what they actually care about and enforcing their own rules back — with nothing leaving the machine. We ship neither the model nor the runtime; Apple does, inside the OS. chanio just points the loop at it. Not a simulation of Apple Intelligence. Apple Intelligence, running the loop.
~/.chanio/memory-pack.md · ~5 KB
# Who you are working with
# What you care about
curated + derived from your usage
# Where you are right now
# Open threads not to lose
# How to help
→ the on-device model loads this
as its instructions. That's it.“I care about building resilient infrastructure that silently fails rather than loudly breaks.”
Not typed into a form. Inferred by the on-device model from what the person actually returned to, across their sessions — then handed back for them to keep or promote.
The consumer version — the same system on the iPhone in your pocket, via the Apple Intelligence primitives above — ships with iOS 27. Same principle, same files, same vow: your memory, your device, your passions moving forward while you sleep.