Why We Built Typilot Local-First
The argument behind Typilot: the AI that learns the most about you should run on the machine you control. Why meetings made cloud indefensible, what local-first buys you, and the bet underneath it.
Typilot is local-first for one reason: the AI that learns the most about you should live on the machine you control, not on someone else's server. Everything else - the privacy guarantees, the lifetime license, the offline mode - follows from that single decision.
This is the argument behind it.
The thing cloud AI quietly asks for
Every cloud AI assistant runs on the same trade: you hand over your words, and in return you get help with them. For a one-off question that is a fair deal. But an assistant that lives on your keyboard does not see one-off questions. It sees everything - the half-written email, the contract clause, the meeting you would never forward, the idea you mumbled into a voice note at 6am.
The more useful that assistant becomes, the more of you it accumulates. With a cloud tool, every bit of that accumulation lands in a database you do not own, governed by a policy that can change, exposed to a breach you will read about after the fact. The better the product gets, the worse the exposure.
We did not want to build a product whose success made its users more vulnerable.
The moment it stopped being optional
Dictation and text polishing are sensitive enough. Meetings are the line. The moment we added meeting recap, "send it to the cloud" stopped being a defensible default. Legal calls, medical consults, M&A discussions, therapy sessions, journalism with a source on the line - uploading that audio to a vendor is not a feature, it is a liability. The people who need meeting AI the most are exactly the ones who cannot use a cloud version of it.
So the recap, the transcription, the diarization, and the chat all run on your machine. Not as a premium tier. As the only way it works.
What local-first actually buys you
Running locally is not just a privacy posture - it changes the economics and the guarantees:
- Privacy by architecture, not by promise. There is no server holding your prompts, so there is nothing to leak, subpoena, or quietly repurpose. You do not have to trust a policy; there is no data to govern.
- No subscription floor. Local inference has a marginal cost of zero. That is why we can offer a one-time lifetime license at all - cloud competitors charge by the minute because they pay by the minute.
- It works in airplane mode. On a plane, in a vault, on a train with bad signal - if it runs on your machine, it runs.
- Your context is portable. Switch to a smarter open model the week it ships; your local history and preferences come with you. No cloud product can offer that without owning your data.
What it costs us
Local-first is not free to build. We give up the easy cloud levers - server-side model upgrades, frictionless telemetry, usage-based pricing that prints money. We ask you to download a model and run Ollama. We optimise hard so a 7B model on a laptop feels instant. That work is the price of the guarantee, and we think the guarantee is the product.
The bet
The bet is simple: as AI gets more personal, people will care more about where it lives. An assistant that has watched you write, decide, and revise for a year is too valuable - and too revealing - to rent from a company that keeps a copy. The future of personal AI is on-device, and we would rather build that now than retrofit privacy onto a cloud product later, when the data is already gone.
That is why Typilot runs on your machine, and always will. If you want the full technical proof, the security page has the architecture, the permissions, and the audits. If you want to see where this goes, read the vision.
Common questions.
What does local-first AI mean?+
Local-first AI means the models run on your own device rather than on a vendor's servers. Your prompts, audio, and transcripts are processed on your machine and never uploaded, so privacy is guaranteed by architecture rather than by a policy.
Why does Typilot run on-device instead of the cloud?+
Because an assistant that lives on your keyboard accumulates everything you write and say. Keeping that on your own machine means there is no server to leak, subpoena, or repurpose it - and local inference has no per-minute cost, which is what makes a one-time lifetime license possible.