Why local-first, in the age of AI
The conventional wisdom is that AI belongs in the cloud — big models, big GPUs, big data lakes. So why run AI — and the data it works on — entirely on your own machine?
Because the interesting AI work isn't the model. It's the operation.
The model is a commodity. Your data isn't.
A capable agent can run the repetitive parts of a business end-to-end: triaging work, advancing deals, screening applicants, drafting documents. But to do that well, it needs reach across your whole operation — and a reason for you to trust what it touched.
That's hard when your data is scattered across a dozen clouds you don't control, each with its own API, its own export limits, and its own terms of service. Every integration is a permission slip. Your business stops being one body of data and becomes a slice rented back to you by each tool you pay.
Local-first changes the calculus
When the interfaces and the database live on your machine — one local file you own:
- An agent can see the whole operation at once, because it's all one data layer.
- You can audit exactly what it did, because the record is yours and every change passes a ship/verify gate where a different agent signs off.
- Nothing leaves your control unless you decide it should. Your model's key is sealed on your machine; the agents bill to your account, not ours.
The AI runs against your data. Your data doesn't run off to the AI.
Ownership is the feature
The defining choice of the SaaS era was to put your data on someone else's computer and rent it back to you. Excellent makes the opposite choice — not out of nostalgia, but because the agent era rewards it. The companies that own their stack will move faster and trust their automation more than the ones renting access to their own information.
It's not a thesis you have to take on faith anymore. It's a file on your disk, and you can run it today.
Windows and Linux are next.