Cue
Your desktop voice agent. Press a hotkey and speak. Cue can type in a focused field, transcribe meetings, use your screen as context, and help finish the task in the app you are using — hands-free, in supported apps.
Office Agent puts the agent where the work is rather than in a chat window: Cue inside the app you already have open, Cue Office inside its own documents. AI Office is the products that read, write and convert documents: PDF AI, PPT AI, a scanner and a book-summary tool.
Installs across every listing on our store accounts.
Ratings people have left on our products.
Their average, weighted by how many rated each.
Dictation-polish latency, down from 876 ms.
Store figures as of 13 September 2026; installs are published in buckets, so the first is a floor. The latency figure is Google DeepMind’s, published with its measurement conditions stated. The first three cover every listing on our store accounts, which is more than the six on the products page: that page shows the products we actively develop, the figures count everything published.
Every one below resolves to a listing you can open today, with the store figures as they read on 13 September 2026.
Your desktop voice agent. Press a hotkey and speak. Cue can type in a focused field, transcribe meetings, use your screen as context, and help finish the task in the app you are using — hands-free, in supported apps.
Docs, Sheets, Slides and PDF in one app, opening and saving .docx, .xlsx, .pptx and .pdf. Ships in 46 languages.



Turn any document into a podcast, slides, or visual digest. Podcast-style audio, AI slides, visual digest cards, and grounded Q&A — as described on Google Play.
Describe the topic and who it is for. You get a structured first draft to edit, not a blank page. Download as PPTX or PDF, or share a link. No lock-in to our editor.



Camera to PDF. Scan a page, crop it, and share the file.



A book, condensed to what you can read in fifteen minutes.
The same four beats a document workflow has always had, now that the agent runs on the desktop rather than on a file.
Wherever you are. Cue does not open a window of its own, so nothing you were doing moves.
Raw speech-to-text arrives with fillers, missing punctuation and self-corrections still in it.
Gemma 4 E4B runs on your own machine through Ollama, which is what took the median from 876 ms to 488 ms.
Pasted through native OS APIs into the field you were already in, in supported apps.
Three claims, each with the address that settles it. Nothing here is graded by us.
Their words: “a voice-activated AI agent that lives on the user’s desktop.” It reports median dictation-polish latency falling from 876 ms to 488 ms once the polish step ran locally on Gemma 4 E4B.
deepmind.google · GemmaverseCounted in the page’s JSON-LD on 14 September 2026, not a number we are asking you to take on trust.
cueoffice.ai · JSON-LD inLanguageWith the project’s own framing attached: “landscape, not leaderboard”.
doi.org · Zenodo 10.5281/zenodo.22227265Deliberately absent: any figure from the demo mockups on heycue.io, which that page itself labels illustrative. Anything we cannot link is left off, including work we have talked about elsewhere: with no public address for it, it does not belong on a page whose whole claim is that everything here can be checked.
Six products across macOS, Windows, iOS, Android and the web, built by a team small enough that the person who writes a feature is the person who answers for it in the store listing.
Open roles, how we hire, and the things we have not figured out yet are on one page.
One address, published in the About-the-developer block of each of our Google Play listings. We read it.