Google just quietly dropped something interesting in the App Store: AI Edge Eloquent, an on-device voice dictation app that works entirely offline, requires no subscription, and has no usage caps. It sits somewhere between an experiment and a consumer product — but it points directly at where AI is heading.
This is not Google’s usual product launch. There was no keynote, no I/O announcement. The app appeared in the iOS App Store in early April 2026 with minimal fanfare. For a company that has been synonymous with cloud AI, releasing a fully offline AI app is a notable signal.
What Is Google AI Edge Eloquent?
AI Edge Eloquent is a voice-to-text app powered by on-device AI. You speak, it transcribes, it cleans up the text, and copies it to your clipboard. That’s the core loop.
What makes it different from the built-in iOS dictation or Google’s own Gboard dictation:
- Runs entirely on your device — no audio sent to the cloud unless you opt in
- No subscription — Google explicitly pitches it as “voice dictation without subscriptions”
- No usage caps — dictate as much as you want
- Text refinement built in — not just raw transcription; it polishes what you said
The app is currently iOS-only. No Android release has been announced.
Key Features
Real-Time Transcription with Live Waveforms
As you speak, the app shows a live waveform and transcribes in real time. You can pause during a session and resume — useful for longer dictation workflows where you need to think between sentences.
Text Refinement After Recording
After you finish speaking, Eloquent gives you post-processing options:
- Key Points — converts your dictation into bullet points
- Formal — rewrites the text in a more professional register
- Short — condenses what you said
- Long — expands and fills out the content
These are simple transformations, but they make the app genuinely useful rather than just a transcription tool.
Fully Offline Mode
There’s an explicit toggle to enable “fully offline” mode. When on, the app guarantees that conversations don’t leave your device. All AI processing — transcription, cleaning, reformatting — happens locally using Google’s on-device model stack.
This is the standout feature. For anything sensitive — medical notes, legal dictation, personal journaling, confidential business communication — fully offline mode means no data exposure risk at all.
Optional Gemini Integration
If you want more powerful text polishing, you can optionally enable Gemini integration. This sends your transcription to Google’s cloud model for enhanced processing. It’s opt-in, clearly separated from the offline mode, and you stay in control of when cloud processing happens.
Custom Dictionary
You can add names, technical terms, and jargon to a custom dictionary that improves transcription accuracy for your specific vocabulary. The app can optionally sync names from your Gmail contacts to pre-populate common names automatically.
Usage Statistics
Each session shows your word count and words-per-minute rate. Small feature, but useful if you’re using this for productivity tracking or just curious how fast you actually dictate.
Why This Matters
On-Device AI Is No Longer a Compromise
The assumption for years was that on-device AI meant worse AI. Smaller models, lower accuracy, fewer capabilities. The tradeoff was privacy and offline access vs. quality.
Eloquent challenges that assumption. Google’s on-device model stack — built through the AI Edge project — is capable enough to power a real consumer app with polished results, not just a basic proof-of-concept. Transcription quality, real-time performance, and text reformatting all work without a network connection.
The Privacy Argument Gets Concrete
Privacy-first AI has been a talking point for a while. Eloquent makes it a product reality. When a company the size of Google ships an app whose headline feature is “we don’t send your voice anywhere,” it validates the demand for on-device AI from a mainstream audience, not just privacy enthusiasts.
For professionals in regulated industries — healthcare, law, finance — this matters immediately. Dictating notes into a cloud-connected app has always carried compliance risk. An app that processes everything locally removes that risk.
No Subscription Is a Positioning Statement
Google pricing Eloquent at zero with no usage caps is notable. It undercuts every subscription dictation service (Dragon Anywhere, Otter.ai, etc.) at the product level, not just the price level. If the quality holds up at scale, it puts real pressure on the paid dictation market.
Who It’s Built For
Writers and journalists — Long-form dictation with polishing tools. Useful for first drafts when you think faster than you type.
Professionals with sensitive content — Lawyers, doctors, therapists, consultants. Offline mode means no compliance headaches.
People who hate typing — Not everyone wants to write every message. Dictate, refine, paste. Three steps instead of one long typing session.
Developers testing edge AI — If you’re building on-device AI apps, Eloquent shows what Google’s own stack can do in a real consumer context. The AI Edge project has released model tooling publicly; this app shows the ceiling of what’s achievable today.
Limitations
- iOS only for now. No Android release timeline.
- Clipboard workflow — the output copies to clipboard rather than integrating directly into apps. You still need to paste manually.
- Gemini integration requires a Google account and network access, which breaks the offline-first story for users who want the enhanced polishing.
- The app is described as sitting “between an experiment and a user-facing tool” — which suggests features and reliability may still be in flux.
The Bigger Picture: Google’s On-Device AI Strategy
Eloquent is not an isolated product. It’s Google’s AI Edge project made visible to regular users. The AI Edge project has been building a framework for running AI models efficiently on device — phones, tablets, laptops — using standard hardware without custom accelerators.
The models powering Eloquent are descendants of the same work that powers on-device features in Pixel phones (live translate, call screening, real-time captions). What’s new is Google packaging it as a standalone app for iOS, putting it in the App Store, and making offline capability the lead feature rather than a footnote.
This suggests Google is testing consumer appetite for on-device AI products before committing to a full platform strategy. If Eloquent gets traction, expect more apps from the AI Edge team — and eventually integration into Android and Chrome OS at the OS level.
How to Get It
Search “AI Edge Eloquent” in the iOS App Store. Install, grant microphone access, and you can dictate immediately without signing in. Enabling Gemini integration requires a Google account.
No subscription prompt. No paywall. No usage limit on the offline features.
Google releasing a privacy-first, offline AI app on iOS is not the headline you’d have predicted a couple of years ago. But it reflects something real: on-device AI has matured enough to be a product, not just a research demo. Eloquent is worth trying if you dictate anything regularly — and worth watching as a signal of where consumer AI is heading.
