Building for airplane mode

The rule

For every feature we ship, we ask: does this work in airplane mode?

If the answer is no, the feature does not ship. Not in a degraded form, not with a "you need an internet connection" toast, not with a sad placeholder. It either works without a network or it does not exist in the product.

This is the most useful single design rule we have written down.

What it forces

• Models live on the device. Not in a cache that warms from the cloud, not as a "fallback for offline." On the device, day one.
• State lives on the device. Notes, transcripts, settings, history — all in local SQLite, with file formats we publish.
• Updates are deliberate. When we ship a model upgrade, you opt in. We do not push and we do not silently swap. The version that worked yesterday will work tomorrow.
• Errors say what they actually mean. "Couldn't reach our servers" is impossible — there are no servers to reach. Errors are local: out of memory, model file missing, mic permission denied. They are honest, and they have a fix you can do yourself.

What it costs

• Cross-device features need pairing dance, not just login. We described that in our other piece.
• No "log in to access your stuff on a new device." A new device is a new device, with a fresh model file, until you pair it.
• No leaderboards, no community feeds, no real-time multi-user. These would all require a network. None of them are part of the products we want to make.

Why this rule, today

Three reasons:

1. It is finally possible. Five years ago, this rule would have meant "ship a much weaker app." Today, with quantized 1–7B models, it means "ship a slightly more focused app." The capability gap is small enough that the trade is worth it.
2. It clarifies decision-making. When the rule is "every feature works offline," the product roadmap writes itself in a way that does not depend on chasing whatever the cloud-vendor de-jour is hyping.
3. It is what users actually want from this category. People do not buy a notebook because it has a great cloud strategy. They buy a notebook because it works in their meetings, on their commute, in their bedroom at 1 AM.

We are betting our company that this is the right rule for the next decade of personal AI. We will let you know how it goes.