- Flip one toggle in Privacy Center and every AI feature — write, summarize, translate, security scan — runs locally through Chrome's bundled Gemini Nano.
- No data leaving your machine.
- Here's what just shipped, how it works, and why it matters for privacy-first email.
TL;DR
Atomic Mail just became the first mainstream email product to run every AI feature through Chrome's built-in Gemini Nano — the on-device LLM Google has been shipping to every Chrome Desktop user since Chrome 138. Users flip a single Local LLM toggle in Privacy Center → Security Options, and writing, summarizing, translating, and security scanning all execute on the user's CPU/GPU. No server calls. No data sprawl. Zero cost. It's live today for all Chrome Desktop users.
What's new
Until today, every “private” email AI still quietly phoned home — prompts and drafts flowed to a cloud model, got processed, and came back. Atomic Mail just cut the wire. Their AI writer, rewriter, tone-shifter, summarizer, grammar checker, translator, and the “security helper” that flags sensitive drafts now all route through window.ai / the Prompt API inside Chrome, invoking Gemini Nano locally on the device. The announcement was surfaced by TestingCatalog, and the toggle is already present for Chrome Desktop users.
Why it matters
This is a category first. Gmail's Gemini features, Outlook Copilot, even Proton's Scribe still round-trip through a remote model. For regulated industries — legal, healthcare, journalism, crypto/OSS security — that round-trip is a dealbreaker regardless of encryption in transit. Atomic Mail's zero-access architecture already meant the vendor couldn't read your mail; pairing that with on-device inference means the AI can't leak it either, because the prompt never leaves your laptop. It also erases the per-token AI bill: local inference is free compute for both the user and Atomic Mail, which is why the feature is included for free-plan users, not gated behind Plus.
Technical facts
- Model: Gemini Nano — Google's smallest Gemini variant, tuned for laptops and desktops.
- Delivery: Chrome auto-downloads the weights (~4 GB) and exposes them through the built-in AI stack (
LanguageModel/ Prompt API). Since Chrome 138+, this ships to every capable Chrome user by default — roughly 3.7 billion monthly actives. - Execution: 100% local on the user's CPU/GPU. No outbound inference calls.
- Cost: $0 per token for the user and for Atomic Mail.
- Works offline: Once the model is cached, AI drafting and summarizing work on a plane.
- Activation: Privacy Center → Security Options → Local LLM toggle inside Atomic Mail.
- Scope of features: Help Me Write, rewriter, tone shifter, summarizer, grammar checker, translator, security helper.
Comparison vs other AI email stacks
| Dimension | Atomic Mail (Local LLM) | Gmail Gemini | Proton Scribe | Outlook Copilot |
|---|---|---|---|---|
| Inference location | On-device (Chrome) | Google Cloud | Proton servers (EU) | Microsoft Cloud |
| Data leaves device? | No | Yes | Yes | Yes |
| Cost to user | $0 | Workspace AI tier | Paid tier | Copilot Pro $20/mo |
| Works offline | Yes | No | No | No |
| E2E encrypted mailbox | Yes (zero-access) | No | Yes | No |
Use cases
The obvious winners are the teams that never wanted cloud AI in their inbox in the first place — lawyers drafting privileged letters, journalists handling sources, healthcare and finance staff under data-residency rules, OSS security researchers triaging disclosures. But the everyday case is simpler: you're on a train with flaky Wi-Fi and you still want the summarizer to condense a 40-message thread before your next meeting. With Local LLM, it just works. And for Atomic Mail free-tier users, it's the first time a full AI email assistant has been available without a subscription.
Limitations & pricing
Gemini Nano is small. It's great at rewrites, tone shifts, summaries, and short drafts — but it won't match Gemini 2.5 Pro or GPT-4 on long-form reasoning. It's also Chrome Desktop only for now: Firefox, Safari, and mobile Chrome don't expose the Prompt API, so users on those browsers fall back to Atomic Mail's cloud AI (still on the Plus plan). The local model needs roughly 4 GB of disk and a reasonably modern machine — Chrome gates the API on hardware capability. First-run warmup is noticeable. Pricing: the Local LLM mode is free for all Atomic Mail users, including the free plan; cloud AI stays a Plus-plan feature.
What's next
Expect the same pattern to land in Edge (which also ships an on-device Gemini-Nano-class model) and eventually Safari once WebKit exposes a comparable API. Atomic Mail's broader 2025–2026 roadmap still has iOS and macOS native apps, custom domains, dark mode, and expanded email aliases queued up. The bigger story is strategic: as Google pushes Gemini features into Chrome across seven new APAC markets this month, Atomic Mail has found a way to ride that same local-model distribution while keeping the privacy-first positioning its users came for.
Source: TestingCatalog, atomicmail.io, Google Developers Blog.
