Download

Four private AI apps. All on your phone. No cloud, ever.

Translate, transcribe, analyze photos, and check symptoms — without sending a byte to the cloud. Built on Google Gemma 4, running entirely on your Android device.

Download APK See why offline
  • 100% offline
  • No upload, ever
  • Gemma 4 inside

What is Cove

Four apps. One model. Zero cloud.

Cove is a suite of four fully on-device AI apps for Android — translation, voice notes, photo understanding, and health Q&A — all running offline on Gemma 4 E2B through LiteRT-LM. No data leaves your phone, no account is required, and nothing depends on a network. Cove keeps working in airplane mode, abroad without roaming, on flights, in tunnels, and behind captive Wi-Fi — anywhere a cloud AI assistant would simply stop responding.

On-device translation on a Pixel 9 returns a short sentence in roughly 280–410 ms — measured against Google Translate's 400–800 ms cloud round-trip on a strong connection, and 2–8 seconds on a flaky one.

  • 280–410 ms First-token latency on Pixel 9, real device
  • 100% offline Works in airplane mode, on flights, abroad
  • Gemma 4 · 4B On-device via LiteRT-LM
  • 0 bytes Uploaded — no data leaves your phone

One family. Four focused tools.

Every app runs the same private, on-device Gemma engine. No cloud. No telemetry.

Why offline matters

Three reasons your data — and your AI — should stay home.

  • Your data never leaves your device

    Cove processes everything — your photos, voice notes, and translations — locally on your Android device. Your data never leaves your device, which means it can't be intercepted, sold, or used to train someone else's AI. This is private AI by architecture, not by privacy policy.

    Read the security whitepaper

  • Works in airplane mode

    Whether you're 35,000 feet over the Atlantic or in a parking garage with no signal, Cove still works. All four apps run fully offline using Gemma 4 on-device. No internet means no rate limits, no outage screens, and no awkward 'one moment, please' waits.

    See the offline demo

  • No round-trip latency

    When AI runs on your phone, there's no network round-trip. Translations appear as fast as you type. Voice transcription keeps up in real time. The latency tax cloud AI charges — Cove doesn't pay it.

    Compare end-to-end speeds

Cove vs cloud AI — what's the actual difference

Most AI apps live in the cloud. Cove lives on your phone. Here's what that buys you across eight common questions.

Question Cove ChatGPT / Gemini Otter / DeepL
Where your data lives On your phone On their servers On their servers
Internet required No Yes Yes
Monthly Pro fee $X.XX one-time* $20+ $17+
Used to train AI? Never Varies by plan¹ Varies by plan¹
Works on a plane Yes No No
AI model Gemma 4 E2B (open weights) GPT-4o / Gemini Closed proprietary
Open weights Yes No No
Voice transcription Local Uploaded Uploaded

¹ Subject to current plan settings; verify in each provider's terms of service.

* See pricing section below for the current Cove all-apps total.

Read the full Cove vs cloud AI comparison

Powered by Google Gemma 4

Open-weight AI, running entirely on your phone

Cove runs on Google Gemma 4 — Google DeepMind's open-weight on-device model. Two variants are available: E2B for fast everyday use, or E4B for higher accuracy on long technical text and rare languages. One model file, shared across all four Cove apps.

  • Gemma 4 E2B (default) ~2-billion effective parameters, ~2.5 GB after quantization. Default download. Smaller, faster — 1–3 seconds per run on mid-range NPUs (Snapdragon 8 Gen 1+, Tensor G2+, Dimensity 9000+). Enough for nearly all travel translation, voice notes, and photo recognition.
  • Gemma 4 E4B (optional upgrade) ~4-billion effective parameters, ~3.6 GB after quantization. Optional download. Higher accuracy on long technical text, rare languages, and multi-turn Q&A. Best on flagship NPUs with 8 GB+ RAM — 1.5–4 seconds per run.

Both variants run on-device through LiteRT-LM, Q4_0 quantized — E2B fits in ~2.5 GB, E4B in ~3.6 GB. Model traffic never leaves your phone. License is the Google Gemma Terms of Use, with weights publicly auditable.

Read the Gemma security whitepaper

Pricing

Pay once. Yours forever.

A single purchase unlocks all four Pro apps — no subscriptions, no renewals.

$12.96 all four Pro apps, lifetime

Otter.ai

$17 /mo

Every month, forever

Cove

$12.96 once

One purchase, yours for life

Otter.ai charges $17 every month. Cove charges once.

Indie-built. No venture money. No dark patterns.

Built for the people who need privacy most

Four real-world scenarios where on-device AI is not a feature — it's the only way the workflow works.

Common questions

Five quick answers. The full FAQ has 30+ more.

  • Does Cove really work without internet?

    Yes. All four Cove apps run fully on-device using Gemma 4. Translation, voice transcription, photo analysis, and symptom check all work in airplane mode. The only network calls Cove makes are to download the model file once on first launch.

  • Is Cove really free to try?

    Yes. Every Cove app has a free daily allowance (5 AI runs per app per day) generous enough to evaluate it for real use. Pro is a one-time purchase per app, not a subscription, and unlocks unlimited daily usage on the same Gemma 4 E2B model.

  • Which AI model does Cove use?

    Cove runs on Google Gemma 4 — Google DeepMind's open-weight on-device model, executed through LiteRT-LM. Two variants are available: E2B (default, ~2.5 GB) covers everyday translation and recognition; E4B (optional upgrade, ~3.6 GB) is more accurate on long text and rare languages. Free and Pro share the same model file; Pro just removes the daily run cap. No cloud inference, ever.

  • Is Cove safer than ChatGPT for sensitive data?

    By architecture, yes. ChatGPT routes every prompt to OpenAI's servers and may use it to train future models depending on your plan. Cove processes everything on your phone — your prompts, photos, and voice memos never leave the device. For travel logs, medical questions, or work notes you wouldn't paste into a public chatbot, Cove is the structurally private choice.

  • Why offline AI instead of cloud?

    Three reasons we built it this way: privacy (your data never leaves the device), reliability (it works on a plane, in a tunnel, behind captive Wi-Fi), and latency (no network round-trip means translations appear as fast as you type). Cloud AI is convenient when it works; on-device AI works when it has to.

Browse all 30+ Cove questions