Local Talk
by localhost/AI
The familiar conversational interface, running an open model on your own hardware. Same shape as the assistants you already use, minus the server.
Not yet released
Announced but not yet on the stores. Links will be added here when it ships.
Local Talk is the plainest expression of the local-first argument: take the interface everyone already understands — a text box, a conversation, a model that answers — and remove the part where the conversation travels to a data center.
What you give up is the frontier. A model small enough to run on a phone is not going to match a hundred-billion-parameter model served from a rack. What you get back is that the conversation is only ever between you and a program on your device. There is no retention policy, because there is nothing to retain. There is no outage, because there is no service. It works on a plane.
For a substantial share of what people actually ask an assistant — drafting, rephrasing, explaining, brainstorming, summarizing something in front of them — a small local model is enough, and the tradeoff is worth making.
What it uses
Text generation. That is the whole app.
Status
In development at localhost/AI. Planned for iOS and Google Play.