Privacy · 5 min read

On-Device AI Translation: Why Privacy-First Email Translation Matters

Most "free" translation tools work by sending your text to a server before showing you a result. For a public tweet, that's harmless. For a client contract, an HR complaint, or a financial negotiation, it's a real exposure — your confidential text now exists on a third party's infrastructure, subject to their retention and logging policies.

What "on-device" translation actually means

Inbox Translate runs a compact, open translation model directly inside your browser using WebAssembly and ONNX Runtime — the same technology that powers modern browser-based AI features. For supported language pairs, your email text is translated entirely on your machine. It never leaves your device, never touches our servers, and isn't logged anywhere.

When does it fall back to the cloud?

Not every language pair has a compact on-device model available yet. When one doesn't, Inbox Translate automatically falls back to a secure backend translation provider (DeepL or Azure Translator) over an encrypted connection — the same standard used by every reputable translation API. You're never left without a translation; you just get the most private option available for that specific language pair.

Our fallback order, in plain terms

• 1. Your device's built-in AI (where the browser supports it)
• 2. On-device ONNX model — runs locally, nothing sent anywhere
• 3. Cached model via CDN — for networks that block model downloads
• 4. Secure backend API — used only when no on-device option exists for that language pair

Why this matters for teams handling sensitive email

If your organization handles legal, medical, or financial correspondence, minimizing what leaves the device isn't a nice-to-have — it's often a compliance requirement. On-device translation gives you a default that's private by construction, not just private by policy.