What is local-first AI?
Local-first AI runs models on hardware you control, and treats the cloud as an option rather than the default. Your prompts and data stay on the machine unless you decide otherwise.
What it means
Most AI tools send everything to a remote service by default. Local-first flips that. The model runs on your own machine, the data stays there, and going to the cloud is a choice you make for a specific task, not a condition of using the tool at all.
Why it matters
Three reasons, mostly. Your prompts do not have to leave your laptop for someone else to read, which is the whole of the privacy argument. There is no meter on local work, so the cost and the rate limits disappear. And it keeps working with the internet off, because the model is already there.
There is a compliance angle too. If the data never moves, most of what the GDPR asks for is already true. We wrote that up in the GDPR docs.
Are local models good enough?
For a large share of everyday work, yes, once you know which model to ask. Code, summaries, translation, transcription, first drafts. The trick is matching the task to a model that handles it well, which is exactly what routing is for. When something genuinely needs a frontier model, you reach for the cloud on that task and stay local for the rest.
Questions about local-first AI
What does local-first AI mean?
It means models run on hardware you control and your data stays there by default. The cloud is opt-in per task rather than the default path, so prompts and completions do not leave the machine unless you choose to send them.
Is local AI good enough for real work?
For a large share of everyday tasks, yes: code, summaries, translation, transcription, first drafts. The key is routing each task to a local model that handles it well, and reaching for the cloud only when a task genuinely needs a frontier model.
Is local-first AI really free?
Running models locally has no per-token cost and no rate limits, and Soriku’s desktop app is free for solo use. You only pay a provider when you choose to route a task to the cloud with your own key.
How does local-first help with GDPR?
If prompts, completions, and agent memory stay on your machine, there is no third-party processing or data transfer to account for in most cases. That makes the default state a compliant one rather than something you configure.