Services - AI features a product can stand behind
We put language models where they help and leave the rest to tested code. In Prandial, the model reads the menu and Swift does the math. In EvenBetter, an agent follows sourced skills to audit an app’s accessibility.
What the work covers
- On the device and in the cloud. Apple Intelligence when the data should stay on the iPhone, Claude when the task needs more, with the routing written in code and a fallback to the device when the network fails.
- Agent skills. Knowledge from a field written as skills for Claude Code and Codex, with steps, sources and outputs that a person checks before deciding.
- Only what the feature needs. Data leaves the device only with consent and only as much as the feature needs, and the privacy policy says exactly what.
- Building with agents. The studio itself builds with Claude Code sessions in the cloud, and the blog shows how that works day to day.
Where you can see it
Prandial
Claude Haiku reads the menu text, Sonnet re-reads what came back thin, and the on-device model takes over offline.
EvenBetter
Open agent skills that audit SwiftUI apps against Apple’s HIG and WCAG 2.2, accepted at JIIHC 2026.
Looté
Explanations from the on-device model, Apple’s cloud or, with Pro, Claude, each shown with its source.
Services - How to work with us
There are two formats. The contact form asks which one you want before it asks about budget.
- Monthly. Development, maintenance and support, month after month. It suits a product that keeps changing after launch.
- Fixed project. Development and support for an agreed scope, without ongoing maintenance. It suits a first version with clear edges.
Questions
Do you train models?
No. We use existing models from Apple and Anthropic and work on what surrounds them: the data, the prompts, the routing and the tests.
What goes to the cloud?
Only what the feature needs. In Prandial, that is the text of the menu, never the photo or the Health data.
Can you build an agent for our team?
Yes, when the work fits a set of skills that a person reviews. EvenBetter is the public example, and its case study shows how it was evaluated.