Four areas of work.

We build the part of an AI product that is not the model: the agent loop, the tool calls, and the permissions around them. Some of that work sits on a server. More of it now sits on the handset.

• Agent runtimes and tool calling
• Model integration across providers
• On-device inference where it fits
• Voice, messaging and automation channels

Native applications, written in Kotlin, Swift and Rust rather than wrapped web pages. We have shipped to Google Play since 2011, which is where most of what we know about batteries, background limits and store review came from.

• Native Android and Wear OS
• macOS and Windows desktop
• Store submission, review and release
• Background work, battery and power budgets

The half of a product nobody sees. APIs, auth, billing, storage, queues and the dashboards an operator runs it from. Mostly Rust and TypeScript, deployed wherever the product already lives.

• APIs and backend services in Rust and TypeScript
• Auth, billing and multi-tenancy
• Operator dashboards and admin tooling
• Telemetry, alerting and release plumbing

The reason for the name. We want the software to sit on the thing it acts on rather than in a datacentre it has to reach, which the industry calls edge computing. The board, the firmware, and what a device still does with no network. We have not shipped hardware and are not selling any.

• Firmware and on-device runtimes
• What a device decides without a network
• Power, thermal and memory budgets