Reticle

Finds and marks what you name, on any screen.
On-device.

Name a control or a piece of content. Reticle locates it and marks it, right where it lives — on any app, any surface, from your exact point of view. A non-invasive AI layer between you and the machine: nothing leaves it, and nothing inside it is touched.

On-device grounding on macOS — finds and marks named content across a live web page and a native app. Press play.

What it is

A non-invasive AI layer between you and your machine — a second pair of eyes for your screen. It closes the gap between what you see and what's actually there — and marks up what it finds on a layer laid over the content, like acetate, never touching what's underneath. It's loyal to you by construction: structurally confined to its own layer, it runs on your machine, reads only the pixels you see — never an app's internals — answers only in its own overlay, and can do only what you could do yourself, in plain sight. And it perceives your screen the way you do, on any app — so it works where tools that read the code underneath come up empty.

The proof under the hood

The Qwen2.5-VL fixes that make on-device grounding reliable in Swift were written by Niv and merged upstream into Apple's mlx-swift-lm.

Merged: #238 · #239 · #242 · #243 · #419  ·  Read the write-up →

Where it reaches that others can't

Native controls, canvas-drawn UIs, remote/VDI desktops — the surfaces where the accessibility tree carries no semantics: it has the words but can't say what is actionable. Reticle works from the pixels and your point of view, so it perceives those surfaces the way you do. It complements the assistive tools you already use — it doesn't replace them — making the parts they can't say are actionable perceivable alongside them.

Get in touch

Worth 20 minutes?

Email me