Meta AI's macOS app arrives with system-wide dictation

Meta AI now has a native macOS client, built around two features that the web version does not have: screen sharing from the prompt bar and system-wide dictation. Meta announced the app on August 19, and it is available in certain regions, where Testing Catalog tested the build.

At a glance

• The prompt bar can take a specific screen or window as an attachment, feeding the contents of that app into the conversation as context, alongside the same functionality Meta AI already offers on the web.
• Voice dictation works through a system-wide shortcut that summons a compact Meta AI prompt bar anywhere in macOS, which puts speech input outside the main application window.
• Artifacts are supported, but the desktop app includes neither browser use nor computer control, and it lacks the deeply integrated coding workflows available to users through ChatGPT and Claude.

A desktop client is where an assistant stops being a browser tab, and Meta has gone without one while its competitors shipped theirs. The macOS release therefore reads as catch-up rather than a lead: screen context and dictation are baseline expectations on that platform, while the absent pieces, browser use and computer control, are what currently separate desktop assistants from web ones. The version 1.0 label suggests Meta treats the product as finished rather than exploratory.

The prompt bar accepts a specific screen or window as context

The client looks close to Meta AI on the web and offers largely the same functionality, with desktop integration accounting for the differences. Screen sharing is the first of them: a user attaches a specific screen or a single application window directly from the prompt bar.

The attached contents then serve as context for the request, which lets the assistant work from what is on screen without a manual description. Testing Catalog documented both screen sharing and the new dictation feature on Threads after the release went out.

The build Testing Catalog accessed is already labelled version 1.0 rather than a beta or preview number. Meta had been preparing a dedicated Meta AI desktop client for macOS before the announcement, and the release turns that work into a shipping product in the regions covered.

Dictation arrives at a company that already ships Omnilingual ASR

Voice dictation is the second capability the desktop client adds on top of the web experience, and Meta named it alongside screen sharing as a core desktop feature when the app was announced. It runs system-wide: a shortcut summons a compact Meta AI prompt bar anywhere on macOS, so dictated input can start without bringing the full application window forward.

Meta already has public speech-recognition technology, including Omnilingual ASR, so dictation does not by itself indicate that a new speech-to-text model is in preparation. Testing Catalog notes that the feature would nonetheless give newer speech technology a consumer-facing surface inside Meta AI.

Muse Code and Muse Spark 1.2 shipped earlier this month

Artifacts are already supported in Meta AI, but the app does not offer the deeply integrated coding workflows available through ChatGPT and Claude. Meta separately launched Muse Code and Muse Spark 1.2 earlier this month, which places the shortfall on the integration side rather than in the underlying coding capability.

Browser use and computer control are absent from the consumer desktop app as well, leaving the client able to read a window a user attaches but not to act inside it. Meta has been experimenting with computer-use capabilities through its developer stack, so that work sits outside the macOS client consumers install for now.

The regions Meta has not named

Meta has not published the list of regions where the macOS app is live, and no timetable for wider availability accompanied the August 19 announcement. Whether the dictation feature will eventually run on newer speech models than the ones Meta has already released publicly is also unresolved, as is any schedule for bringing browser use or computer control into the consumer client.