The On-Device AI Note-Taker Is a Wake-Up Call for Every Seller Who Records Calls
Cross-border sellers live and die on recordings. Supplier negotiations on WhatsApp voice memos. Buyer calls at 2 a.m. your time. A factory visit where the QC manager rattles off tolerances in fast Mandarin while your phone sits in your pocket. The tooling that transcribes and summarizes those recordings is almost always cloud-based, subscription-priced, and quietly uploading your commercial conversations to someone else’s server. So when a 21-year-old computer science student in Germany ships a note-taker that runs Whisper transcription directly on the handset — no account, no ads, no subscription — it’s worth more than a polite Product Hunt upvote. It’s a design pattern your ops stack should steal.
Earnote, built by Louis Klinke, is nominally a study tool. Read it as a study tool and you’ll miss the point. Read it as a proof of concept for private, offline, zero-marginal-cost voice capture, and it becomes a template for the next generation of seller-side tooling — the kind that doesn’t leak your supplier pricing to a SaaS vendor’s training pipeline.
What the Product Actually Does (and Why the Architecture Matters More Than the Features)
The mechanics are simple. You put your phone on the table and record. Or you feed it a voice memo, an interview, or a lecture video. Earnote transcribes with Whisper on the device, then generates a note: a summary, the tasks and deadlines that were actually mentioned, flashcards for self-quizzing, and a PDF study sheet. On iPhone 15 Pro and later, the note generation itself happens on-device, so the whole loop works in airplane mode. There’s no account, no ads, no subscription, and everything core is free. An optional one-time Pro tier on iPhone and iPad adds speaker detection, question-asking over a note, and translation. The Mac app is free and open source under MIT.
For a cross-border operator, the interesting column isn’t “flashcards.” It’s the combination of three properties:
- On-device transcription. The audio never leaves the machine. That matters when the audio is a supplier quoting you a factory-gate price that you’d rather not have sitting in a third party’s object storage.
- No account. No login means no seat management, no SSO procurement, no admin console, no data processing agreement to negotiate. For a five-person sourcing team, that’s a real reduction in legal overhead.
- Zero marginal cost. A one-time Pro purchase versus a per-seat monthly subscription changes the math on how liberally you deploy the tool. When transcription is free at the margin, you record everything. When it’s $0.30 an hour plus a seat fee, you ration it — and you miss things.
Where the math breaks
The “free forever, one-time Pro” model works for a solo student because his marginal cost is genuinely zero — the compute is on the user’s own silicon. That’s also its ceiling. There’s no revenue to fund an Android build, no revenue to fund a server-side pipeline for older devices, no revenue to fund enterprise features like shared team workspaces or audit logs. Klinke is a student shipping a side project, and the pricing reflects that. Don’t expect an SLA.
How It Stacks Up Against What Sellers Actually Use Today
The honest comparison set for a seller isn’t other note-taking apps. It’s the transcription and meeting-intelligence layer that has quietly become standard in cross-border ops.
Otter.ai is the default for English-language meeting capture. It’s cloud-first, account-based, and priced per seat on a subscription. It’s excellent at what it does and completely wrong for the supplier-negotiation use case, because the whole value proposition assumes your audio is fine living on Otter’s servers.
Fireflies.ai and Fathom push further into CRM integration and sales-call summarization. Great if your “meetings” are Zoom calls with US buyers. Useless if your “meetings” are a WeChat voice message from a Shenzhen trading company at 11 p.m.
Whisper itself, the open-source model underneath Earnote, is the real incumbent here. Anyone with a GPU and an afternoon can run it locally. Earnote’s contribution is packaging: the recording UX, the note-generation layer, and the on-device deployment on Apple silicon. If you’re technical, you can replicate 80% of this yourself. If you’re not, Earnote is the shortest path to a private transcription loop.
The comparison that matters most, though, is the phone’s own built-in tools. iOS Voice Memos now transcribes. Android’s Recorder app does too. Both are free, both are on-device on modern hardware, and both are already in your pocket. Earnote’s edge over them is the downstream note generation — the summary, the extracted tasks, the flashcards. Whether that’s worth installing a third-party app depends entirely on whether you actually want structured output or just a transcript.
What Cross-Border Sellers Should Borrow From This
Strip away the study-app framing and there are four patterns here worth copying into your own operations, regardless of whether you ever install Earnote.
Pattern 1: Record the supplier call, on-device, every time
Most sourcing teams take notes by hand during supplier calls and lose the nuance. The price was “$2.80 at 5,000 units, $2.60 at 10,000, and we can talk about tooling amortization.” By the time you’ve written that down, you’ve missed the next sentence. An on-device recorder that transcribes locally and summarizes into action items — whether Earnote, iOS Voice Memos, or a self-hosted Whisper pipeline — turns every supplier call into a searchable artifact without exposing the audio to a vendor. If you’re running sourcing through Alibaba or direct factory relationships, this is the single highest-leverage process change available to you this quarter.
Pattern 2: Treat “no account” as a feature, not a limitation
Procurement teams hate seat-based SaaS for exactly the reason Klinke cites: the subscription. A tool with no account and no per-seat pricing can be deployed to your entire sourcing team, your 3PL contacts, and your freelance VAs without a procurement cycle. When you’re evaluating your own tooling stack — your Shopify app stack, your Amazon Seller Central reporting layer, your Klaviyo flows — ask which of those tools would be better as a one-time-purchase, local-first utility. The answer is more of them than you think.
Pattern 3: The “extract the tasks and deadlines” output is the actual product
Klinke’s differentiator isn’t transcription. It’s that the note contains “the tasks and deadlines that were actually mentioned.” That’s the same problem every seller has with buyer calls, 3PL onboarding calls, and marketplace account manager reviews. A transcript is a haystack. A structured list of commitments — who owes what by when — is the needle. When you build or buy tooling, optimize for the structured output, not the raw capture.
Pattern 4: Open source as a hedge against vendor risk
The Mac app is MIT-licensed. That means if Klinke walks away, the code doesn’t. For a seller building internal tooling on top of a vendor’s API, that’s a meaningful difference. When you’re choosing between a closed SaaS and an open-source equivalent, the open-source option is a hedge against the vendor’s roadmap, pricing, and survival.
Where My Judgment Says It Falls Short
I like the architecture. I’m skeptical of the product as a business, and sellers should be too.
The Android gap is disqualifying for most field ops. Klinke’s own comment thread has a user asking directly about Android, and the honest answer is that the on-device note writing depends on Apple’s silicon. That’s fine for a student with an iPhone 15 Pro. It’s not fine for a sourcing agent in Guangzhou carrying a mid-range Android phone. The majority of the world’s cross-border supply chain runs on Android. A tool that only works on recent Apple hardware is a tool for the buyer side of the desk, not the factory floor.
“Out-loud correction” is unsolved. One commenter raises a genuinely hard problem: you say a wrong date, then correct it a sentence later, and the note doesn’t know which one you meant. This is the exact failure mode that kills transcription tools in commercial use. A supplier says “we can do 30 days” then “actually, 45 for the first order.” If your summary picks 30, you’ve just built a supply chain risk into your notes. Earnote doesn’t claim to solve this, and neither does anyone else convincingly.
No team layer, no audit trail, no compliance story. For a solo operator, fine. For a brand with five sourcing staff and a compliance obligation to retain commercial records, the absence of shared workspaces, retention controls, and audit logs is a real gap. The “no account” feature that’s a strength for privacy is a weakness for governance.
The study-app framing limits the roadmap. Flashcards and PDF study sheets are the wrong output for a seller. What a seller wants is a structured call summary pushed into a Notion database, a Slack channel, or a HubSpot deal record. Earnote doesn’t do that, and given the one-person team and the one-time pricing, it probably won’t.
One-time pricing means no support. If the app breaks on an iOS update the week before your biggest supplier negotiation, you have no recourse. That’s the trade you make for free.
What I’d Watch / Test Next
Three concrete things to do this week, in order of leverage.
First, audit your own call-recording exposure. List every tool in your stack that touches audio or video from supplier or buyer conversations. For each, answer: where does the file live, who can access it, and what’s the retention policy? If the answer to any of those is “I don’t know,” you have a data governance problem that predates Earnote and will outlive it. The on-device pattern Earnote demonstrates is the fix — whether you adopt Earnote or replicate it with a local Whisper install.
Second, test the on-device transcription quality on your actual audio. Record a real supplier call — with consent — in the language and accent you actually deal with. Run it through Earnote on an iPhone 15 Pro if you have one, and through OpenAI’s Whisper locally on a laptop if you don’t. Compare the structured output, not the raw transcript. The question isn’t “is it accurate” but “does the summary catch the commitments.”
Third, watch the Android question. Klinke’s comment thread is the place to track it. If an Android build ships with on-device note generation, this stops being a curiosity and becomes a deployable field tool for sourcing teams. Until then, treat it as a design reference, not a production dependency.
The broader lesson for cross-border operators: the AI tooling wave is moving toward local-first, private-by-default, zero-marginal-cost utilities. The SaaS vendors selling per-seat cloud transcription are on the wrong side of that curve for the most sensitive conversations in your business. Earnote is a student project, but the architecture it demonstrates is the direction the whole category is heading — and the sellers who internalize that early will be the ones whose supplier negotiations stay private.
