How does Nova understand what's on screen?

When context reading is on, Nova reads the text content of the active window — the message you're replying to, the subject, the tone — to write a fitting reply instead of generic text. It reads first through Windows accessibility (light, local), then through local on-device text recognition (OCR) if the app doesn't expose its content. All of it stays on your machine.

Which apps does it work with?

The active window, whatever it is — mail, tickets, documents, chat. Apps that hide their text from accessibility (often Chromium- or Electron-based ones) are covered by local text recognition, which reads what's shown on screen.

Feature

The dictation that reads your screen to reply in context

Q: Does Nova send screenshots?

Not by default. Context reading runs locally: nothing leaves your computer. An optional 'cloud vision' feature, available only on Nova Ultra, can describe a screenshot server-side for the rare windows that are otherwise unreadable — but it's off by default, every send is an explicit choice, and the image is never stored.

Q: Is context reading tied to a subscription?

Yes, context reading is part of Nova Ultra. Local dictation and rewriting remain free.

Q: Does it slow dictation down?

Reading through accessibility is near-instant. Local text recognition only kicks in as a fallback, when an app doesn't expose its content. And context reading is only used to understand the situation: if your dictation stands on its own, Nova ignores it.

Updated July 20, 2026 · 4 min read

Dictating a reply without context is like answering a message you haven't read: the text comes out correct, but generic. Nova changes that — when you allow it, it reads the context on your screen (who you're replying to, about what, in what tone) to write a reply that actually fits. And it does so locally by default: your screen goes nowhere.

The gist in 30 seconds

What it does: Nova understands the situation on screen to write a fitting reply, not a one-size-fits-all one.
Where it runs: locally first — Windows accessibility, then local text recognition as a fallback. Nothing leaves your machine.
The cloud option: Nova Ultra only, off by default, every send is a choice — and the image is never stored.

The problem: dictating blind

Ordinary dictation only hears you. It has no idea the window in front of you holds an unhappy customer, a precise support ticket, or a friend's message in a light tone. So you dictate "tell them it's fine" and get a flat sentence you have to rework so it truly fits the conversation.

What Nova does: read the context, not your privacy

With context reading, Nova looks at the content of the active window as you dictate, and uses it only to understand the situation. It doesn't copy what it reads and doesn't reply on your behalf: it tunes your text so it lands right. That reading follows a privacy-first cascade:

Windows accessibility — Nova reads the text the app already exposes to the system. Light, instant, 100% local.
Local text recognition — if the app hides its content (often Chromium- or Electron-based software), Nova reads what's shown, still on your machine.
Cloud option (Nova Ultra, opt-in) — for the rare windows unreadable any other way, a description can be requested server-side. Off by default, enabled explicitly, image not stored.

The rule: local always goes first. The cloud only comes into play if you've enabled it, and only as a last resort.

Local first, cloud only if you decide

It's the same philosophy as the rest of Nova: dictation and rewriting already work fully offline, no account. Context reading extends that principle. By default, no image and no text from your screen leaves the computer — we can't read them. The day you turn the cloud option on, it's a conscious choice, reserved for cases where local isn't enough, and flagged as such.

What it's actually for

Answer an email without re-reading the whole thread: dictate the idea, Nova matches the tone and details to the message in front of you.
Handle a ticket or support message: the reply picks up the right number, the right name, the right problem.
Write in the right register: formal with a client, direct with a colleague — because Nova sees who you're talking to.

Frequently asked

Does Nova send screenshots of my screen?

Not by default. All context reading happens locally. Only Nova Ultra's "cloud vision" option can describe a screenshot server-side — it's off out of the box, every send is explicit, and the image is never stored.

Is it subscription-only?

Context reading is part of Nova Ultra. Local dictation and rewriting stay free.

Does it slow dictation down?

Reading through accessibility is near-instant; local text recognition only kicks in as a fallback. And if your dictation stands on its own, Nova ignores the context entirely.

Which apps?

The active window, whatever it is. Apps that don't expose their text are covered by local recognition, which reads what's on screen.

Try the dictation that understands the situation.
Hold F9, speak — Nova tunes the reply to what's in front of you.

Download Nova for Windows

Free for local dictation · context reading included in Nova Ultra · Windows 10/11 · see pricing