Frontal is an operations assistant we built for ourselves and run on our own server. It brings mail, payments, renewals, deadlines, projects and system health into one phone app. A local AI model sorts incoming items, drafts replies and flags what needs attention.
Frontal has one user and is not for sale. We show it because it is a useful, working example of how OMWSD approaches AI in software: give it a defined job, keep the result visible, and make it clear where a person remains in control.
One place to review the work
Frontal is organised around a home screen, a queue of decisions, mail, projects, notes and history. There is also a chat that can check live status. The point is to bring related work into one place without pretending that every source has become the same thing.
For example, mail still has its thread and sender. A renewal still has its date. A system alert still points to the system it concerns. Frontal presents those items together so they can be reviewed from a phone, with enough context to decide what happens next.
What the local model does
The model sorts incoming items by importance. When a reply is useful, Frontal can draft one from the message thread, past dealings and the owner's writing style. The draft is there to review and edit. Corrections to the model's sorting become rules that take priority over its next suggestion.
Some repeat replies can be set up as standing orders. They pass through approvals before sending on their own, and a rejection removes that permission. This keeps routine work narrow and reviewable instead of asking a model to make broad decisions without context.
Frontal also re-reads open mail threads overnight. It records a promise only when the quoted words are actually in the thread, then prepares replies for the morning. For system and backup checks, it reads status rather than changing the systems it watches.
“Local AI” has a specific meaning here
The sorting model, speech recognition and spoken brief run on a GPU we operate. Other parts have different boundaries: coding work orders go to an external coding tool, and phone notifications pass through a public relay. Describing the model as local should not imply that every connected operation stays on the same machine.
That distinction is part of the design. A product should say which work happens locally, which services it relies on, and what the model is allowed to do. Those are implementation choices to make deliberately, not a promise that the word “AI” can carry by itself.
A working example, not a product offer
Frontal is built for one person and is not a product people can buy. It is an internal tool that shows how OMWSD combines a phone interface, backend services, local models and ongoing operations in a real workflow.
You can read the Frontal case study for the features and technical details. If the problem you are solving involves a connected web product or its back end, see OMWSD's work in web and SaaS platforms and backend infrastructure.