Sight protection: guarding what the eyes see — not just which sites load

Name the actual object of this community's device standards and it was never the URL. Bamidbar's charge — do not stray after your hearts and after your eyes — names the eyes; the entire edifice of communal practice, from the earliest kosher phones to the certified market of 2026, exists to govern what gets seen. But the tools available for twenty years could not touch seeing itself, so they governed proxies: which addresses load, which apps install, which networks connect. Proxy governance worked exactly as well as the proxies tracked the target — and the tracking has collapsed. Content no longer lives at addresses; it arrives through allowed doors — the message attachment, the shared album, the embedded frame, the search results page itself — and a toolbox aimed at addresses is structurally blind to what walks through an approved one.

The kolbo.life homepage's security section names the post-proxy answer: "AI sight protection — State-of-the-art models screen images, video, and text in real time. Protection at the level of what the eyes see — not just which sites load."

From proxies to the thing itself

Read the sentence's second line as the design revolution it is. "Not just which sites load" retires the proxy era in six words: the unit of governance moves from the address to the content — the image as it renders, the video as it plays, the text as it appears — evaluated in the moment, wherever it came from. The approved door stops being a blind spot because doors stop being the question; the screen's actual output is. This is the standard's original object — the seeing — finally addressable directly, which is why the feature's name is not a technical term but the community's own category translated: sight protection. Shmiras einayim, as architecture.

The mechanism the homepage names — "state-of-the-art models" — is the era's one genuinely new capability put to this community's oldest purpose: machine perception fast enough to evaluate content in real time. This library's AI cluster draws the distinction that matters here: AI-as-assistant is the kind requiring boundaries and deniability; AI-as-guardian — this kind — is the community's standard running at machine speed, and the platform deploys the two with opposite postures on purpose.

Where the layer works

Because sight protection belongs to the security layer — "under everything we build" — it serves surfaces, not just sites, and this library meets it across the whole map. In the browser, where it screens what pages actually render — the difference between a bouncer with a guest list and one who can see. In the kids' browsing question, where "judges what renders" is the structural bar no blocklist reaches. In the Camera, where the homepage places protection in the capture pipeline itself — the lens joining the standard. In the mail cluster's threat picture, where AI-fluent fraud is exactly the content-not-address risk proxies miss. And on any iPhone or Android today, where the layer ships as part of KolBo Secure, from $14.99/month — the guardian available before the suite's devices arrive.

“The proxies — addresses, apps, networks — tracked the target for twenty years. The target was always the seeing, and the seeing is finally the thing guarded.”

kolbo.life

The disclosures, per this library's rule: model architectures, category coverage, accuracy figures, and configuration specifics aren't stated on the homepage — the quoted sentences are the claims, and hello@kolbo.life answers past them. The historical point stands on the homepage's words alone: every generation of this community's device standards aimed at the eyes and settled for proxies, because the direct object was beyond any tool's reach. "Protection at the level of what the eyes see" is the settling, ended.

Frequently asked questions