The Local-AI Question Cross-Border Operators Keep Asking Wrong
Most cross-border sellers I talk to are still treating AI as a browser tab — paste your supplier email into ChatGPT, paste your listing copy into Claude, hope the cloud keeps your margin math and your supplier pricing out of someone else’s training set. That works until it doesn’t: until you’re handling a factory cost sheet you signed an NDA for, a customer dispute file with PII in it, or a competitor-teardown spreadsheet you’d rather not upload to a US server. Gerty, a new macOS app from Swiss maker Erik Vetsch, is worth your attention not because it’s a better chatbot but because it reframes the question: what if the model selection problem — the actual reason most operators never get past ChatGPT — was the thing being automated? That’s a small shift with outsized implications for anyone running a lean DTC or FBA operation across time zones.
The Real Problem Gerty Is Solving (And Why It’s Not the One You Think)
Read the launch copy carefully and the pitch is not “local AI.” It’s “model selection as a service.” Vetsch’s framing is that there are more than two million open AI models in circulation, plenty of them run fine on a Mac, and the hard part is knowing which ones match your specific hardware and task. That’s a real bottleneck. LM Studio and Ollama — the two incumbents anyone in the local-LLM space already knows — lower the barrier to running a model, but you still have to pick it, size it against your RAM, download it, and wire it into whatever workflow you want. Gerty measures your Mac, picks the strongest open models per job category (chat, writing, code, image understanding, image generation), downloads them, and assembles them into a single assistant. You pick a package, it shows you the download size, you ask.
For a cross-border operator, that’s the difference between “I’ll get around to setting up a local LLM someday” and “I have a private assistant on my laptop by Friday.” The tasks Gerty advertises map almost suspiciously well onto seller pain points: ask a folder of documents a question with scans included and see the file and page the answer came from (supplier contracts, freight invoices, compliance certificates); find mail and draft replies in your own writing style (buyer messages, supplier negotiations); read and add calendar events, reminders, and notes and run your Shortcuts (launch calendars, restock reminders); ask out loud from the notch with a locally-running voice; and point Claude Code, Codex, or aider at Gerty’s local API instead of a cloud endpoint. That last one is the sleeper feature. If you’ve got a technical ops person building internal tooling — a repricing script, a listing QA bot — being able to point their coding agent at a local API means the code never leaves the machine.
The honest limits are, well, honest: Apple silicon and macOS 26 required, not open source, and the biggest cloud models still beat it on the hardest reasoning. Vetsch says 16 GB or more handles everyday work well. Pricing is CHF 49 once for up to three Macs, a 7-day trial with no sign-up, and a 14-day refund.
Why Amazon sellers should care more than Shopify ones
Shopify operators live in a browser. Their stack — Shopify admin, Klaviyo, Meta Ads Manager, a dozen SaaS dashboards — is cloud-native by design, and most of their sensitive data is already sitting in someone else’s SOC 2 environment. The marginal value of a local assistant is convenience, not confidentiality.
Amazon FBA brand owners are a different animal. Your Seller Central data, your supplier POs, your inspection reports, your trademark filings, your reimbursement cases — a lot of that is material you’d genuinely rather not paste into a consumer chatbot, both for competitive reasons and because Amazon’s own terms around data handling are strict enough that sloppy AI hygiene is a real account-health risk. A local assistant that can read a folder of scanned inspection PDFs and cite the page is, for this cohort, closer to a compliance tool than a productivity toy. Same logic applies to Temu and SHEIN sellers dealing with factory documentation in Chinese, though Gerty’s model lineup — Gemma-based, per the maker’s reply — will need to prove itself on CJK document OCR before I’d trust it there.
How It Stacks Up Against What You’re Probably Already Using
The fair comparison set isn’t ChatGPT vs. Gerty. It’s three tiers.
Tier one: cloud assistants. ChatGPT, Claude, Gemini. Better at hard reasoning, zero setup, and the ones your team already uses. Gerty doesn’t try to beat them there — Vetsch concedes the point in his own launch copy. What Gerty offers instead is the offline, no-account, no-telemetry posture and the ability to work with files you’d never upload.
Tier two: local runtimes. LM Studio and Ollama. Both are excellent and both put the model-selection burden on you. If you have an ops engineer who enjoys that, fine. If you’re a founder-operator with forty SKUs and a three-person team, “enjoy” is doing a lot of work in that sentence. Gerty’s whole wedge is collapsing that decision tree.
Tier three: vertical AI tools. Helium 10’s AI features, Jasper for copy, whatever repricing SaaS you’re on. These are task-specific and cloud-hosted. Gerty is general-purpose and local. They’re complementary, not competitive — Gerty won’t replace your listing optimizer, but it might replace the ad-hoc ChatGPT tab you use for everything the vertical tools don’t cover.
The interesting positioning is that Gerty sits between tiers two and three: more opinionated than a raw runtime, more general than a vertical tool, and local in a way neither of the others is.
Where the math breaks
CHF 49 for three Macs is cheap enough that the price isn’t the objection. The objections are elsewhere. First, the 18 GB first download for Gerty Max on a 16 GB Mac — that’s the maker’s own number from the comments — is a real commitment on a laptop you’re also using for actual work. Second, macOS 26 and Apple silicon means anyone on an Intel Mac or a Windows machine is out, which in a cross-border team usually means the warehouse ops lead and the China-based sourcing person are out. Third, “not open source” plus “no telemetry” is a trust posture, not a verifiable one — you’re taking the maker’s word for it. For a solo founder that’s fine. For a brand with a legal team, it’s a conversation.
What Cross-Border Sellers Should Actually Borrow From This
Three transferable lessons, independent of whether you buy Gerty.
One: the “honest limits” section is a conversion asset, not a liability. Vetsch leads with what the product can’t do — needs Apple silicon, not open source, weaker than cloud on hard reasoning — and the top comment on the launch is literally praise for that section. Cross-border sellers write product pages that read like every competitor’s: five bullets, all superlatives, no acknowledgment of fit. The operators winning on Etsy and eBay right now are the ones whose listings say “not for X use case” up front, because it kills the returns before they happen. Steal this. Your A+ content and your Shopify PDP should have an honest-limits block.
Two: bundle the setup, not just the tool. Gerty’s insight is that the configuration is the product. Sellers do this instinctively with kits and bundles — the reason a “starter pack” outsells its components — but rarely with their own operations. If you’re selling a complex SKU, the unboxing and first-use setup is your Gerty moment. Invest there.
Three: local-first is a positioning wedge in a cloud-saturated market. Whether it’s AI or anything else, “your data stays here” is becoming a premium feature again, the way “made in USA” cycled back. For sellers in regulated categories — supplements, cosmetics, children’s products — a supplier or 3PL that can credibly say “your formulations and test reports never leave our facility” is worth a margin point.
Where My Judgment Says It Falls Short
I like the product’s thesis more than I like its current execution for this audience.
The model bundle structure — Gerty Max vs. Gerty Thinking, per the maker’s reply to a commenter — is sensible but under-explained on the launch page. A seller who only wants document Q&A shouldn’t have to reverse-engineer which bundle skips the image models. The 330 MB semantic-search model that downloads “the first time you use it” is exactly the kind of surprise download that erodes trust in a metered-bandwidth environment; ship it in the base bundle or warn louder.
More substantively: the seller use cases Gerty lists are generic knowledge-worker tasks. Nothing in the launch copy speaks to the actual workflows that eat a cross-border operator’s day — parsing a supplier’s WeChat export, reconciling a freight forwarder’s invoice against a PO, drafting a Section 321 or IOSS-friendly customs description, auditing a listing against Amazon’s restricted-claims policy. That’s not a knock on Gerty specifically; it’s a gap in the whole local-AI category. The first product that ships a “cross-border seller” bundle — pre-loaded prompts, document templates, a supplier-email parser — will own this niche. Vetsch explicitly asks the community which packages they’d want covered, so there’s an opening to tell him.
And the platform lock is real. A three-Mac license sounds generous until you realize the person who most needs a private assistant is often the one on a Windows ThinkPad in Shenzhen. Not disclosed whether a Windows build is planned.
What I’d Watch / Test Next
This week, three concrete moves.
First, if you’re on Apple silicon, take the no-sign-up 7-day trial and run one real task through it: point Gerty at a folder of five supplier PDFs — a PO, an inspection report, a freight invoice, a compliance cert, a packing list — and ask it to reconcile quantities across all five and cite the page for each discrepancy. That’s the single highest-value test for a cross-border operator, and it’s the kind of task where a local model’s page-citation feature either earns its keep or doesn’t.
Second, regardless of whether you adopt Gerty, audit your team’s current AI hygiene. List every workflow where someone pastes supplier pricing, customer PII, or account-health data into a cloud chatbot. That list is your local-AI business case, and it’s also your next SOP revision.
Third, watch the comment thread on the Product Hunt launch for the next week. The maker is answering bundle questions in real time, and the requests that pile up will tell you whether a seller-specific package is coming. If it does, that’s the version worth paying CHF 49 for. If it doesn’t, you’ve learned something about the category’s blind spot — and about where the next wedge for a cross-border-native tool is hiding.
