Why Local-First Matters

There's a common argument for cloud software that goes like this: your data is safer on our servers than on your laptop. We have redundancies, backups, and security teams. Your laptop could get stolen, broken, or wiped.

This argument is reasonable for some kinds of data. It is not reasonable for all data. And it conflates two different things: the durability of your data and the sovereignty of your data.

Durability is about not losing things. Backups, replication, and redundancy solve for durability. Cloud services are genuinely good at this.

Sovereignty is about who controls your data. Where it lives. Who can access it. Whether it continues to be available if the company changes its terms, gets acquired, or shuts down. Cloud services are, by definition, a compromise on sovereignty. Your data lives on someone else's computer. You are trusting them.

For many people, this trade-off is acceptable. The convenience of cloud sync, collaboration features, and zero-maintenance infrastructure outweighs the sovereignty cost. But for certain kinds of work — and certain kinds of data — the sovereignty cost is too high.

Voice dictation is one of those cases. When you dictate, you're not just producing text. You're externalizing thought in its rawest form. Your voice carries hesitation, emotion, accent, and cadence. It's personal in a way that typed text isn't. Sending it to a cloud service — even a well-intentioned one — means giving up control over something intimate.

This is why Steno processes everything locally. The speech-to-text model runs on your Mac's Neural Engine. The optional refinement model runs on the same chip. Your voice never leaves your machine. Not for processing. Not for storage. Not for "improving the service."

Local-first isn't just about privacy, though. It's also about reliability. When your software depends on a cloud service, you introduce a dependency chain: your internet connection, the service's uptime, the service's API stability, the service's continued existence. When a cloud dictation service goes down — or worse, shuts down — your workflow breaks.

Local software doesn't have this problem. It works the same way at a café with spotty WiFi as it does at a desk with gigabit fiber. It works on a plane. It works during an outage. It works because the software and the data are both on your machine, and no external dependency can take them away.

There's also a longevity argument for local-first software. Cloud services die. Companies pivot, get acquired, or run out of money. When a cloud service shuts down, your data and your workflow go with it. Local software, once installed, continues to work. The binaries don't expire. The models don't get deprecated. The tool remains yours.

This doesn't mean there's no place for cloud software. Collaboration, sync across devices, and access to powerful compute are genuine benefits that local software struggles to match. But the pendulum has swung too far toward "everything in the cloud." We've accepted dependence on remote services for things that work perfectly well — often better — on the machine in front of us.

Local-first is a corrective. It's a reminder that not everything needs to live on someone else's server. That the fastest, most private, most reliable place to run software is often the device already in your hands.

We're building for the people who understand this trade-off and choose sovereignty. Not because they have something to hide. Because they have something to protect.