The Exit Clause Is the Product
Cross-border operators don’t lose money to bad software. We lose it to software we can’t leave. A marketplace account gets suspended and the tooling around it becomes worthless overnight. A note-taking app gets acquired, the price doubles, the free tier shrinks to fifty notes, and fifteen years of supplier notes, customs paperwork, and product research sit behind an export button that technically exists and practically doesn’t. So when Harbor launched on Product Hunt this week with a founder story built entirely around that exact betrayal, I read it less as a note-taking app review and more as a case study in the exit-clause economics every DTC operator should be running against their own stack. Spicer Matthews, founder of Cloudmanic Labs, spent over a decade inside Evernote, watched it get sold, watched the price roughly double, watched the free plan shrink to fifty notes, and watched his own data get harder to extract. Harbor is his answer. Whether you need another second brain is beside the point. Whether your current stack would let you rebuild in a week is not.
What Harbor Actually Solves, and Why It’s Not a Note App
Strip the positioning away and Harbor is a personal knowledge store with three load-bearing features: search that reads the words inside photos, scanned PDFs, and handwriting; recordings that become transcripts; and a full export on every plan in HTML, Markdown, and ENEX, plus an open API, a CLI, and an MCP server.
That third one is the actual product. Everything else is table stakes in 2025.
Matthews is explicit that the whole thing was built in reaction to a specific failure mode: the app got better while the deal got worse. He counted Evernote’s 500 most recent US App Store reviews and published the breakdown. Of the negative reviews, 9% mention a bug, crash, or sync problem. 54% are about price. Read that ratio again. Users weren’t leaving because the software broke. They were leaving because the terms moved underneath them. That’s a distinction every SaaS-dependent operator should internalize, because it describes almost every tool in a cross-border stack. Your Shopify plan, your Klaviyo tier, your Helium 10 seat count — none of them are failing you technically. They’re repricing you.
Why Amazon sellers should care more than Shopify ones
Shopify merchants have a real exit. You can migrate a storefront to WooCommerce or BigCommerce, and while it hurts, the data model is portable enough that agencies do it for a living. Amazon sellers have no such luxury. Your listing content, your A+ modules, your Brand Analytics search-term reports, your advertising history — none of it leaves Amazon Seller Central in a form you can rebuild from. You are a tenant, not an owner. The Harbor thesis lands harder the deeper you sit inside a walled garden, and there is no deeper walled garden in commerce than Amazon.
Which is why the “bring your own AI” decision is more interesting than it sounds. Harbor ships with no built-in assistant. Instead it exposes an MCP server and lets you connect the ChatGPT or Claude subscription you already pay for. For an operator, that’s the correct architecture for a boring reason: you don’t want a second AI vendor with a second billing relationship and a second data-processing agreement. You want one model contract, one set of prompts, and a retrieval layer that any of them can query. The tool becomes a database. The intelligence stays swappable.
The Pricing Promise Is the Most Interesting Thing Here
One plan. $9.99/mo or $99/yr, free to start. No increase for three years after you subscribe, then capped at 10% a year. Stop paying and the account goes read-only, never deleted. And the promise is written into every new account as a note, dated the day you joined.
That last detail is the one I keep coming back to. It’s not a marketing page. It’s a timestamped artifact inside your own workspace, which means it survives a landing-page redesign, a pricing-page A/B test, and a new head of growth who’s never read the original blog post. Matthews is essentially saying: the terms are data, and data you control is data that can’t be quietly edited. For anyone who has watched a vendor’s “grandfathered” plan get un-grandfathered, this is a genuinely novel enforcement mechanism. Not legally bulletproof — a note in your account is not a contract — but socially expensive to break in public.
The read-only-on-nonpayment clause deserves separate applause. Most tools in this category either delete your data or hold it hostage behind a reactivation paywall. Read-only is the correct middle: you can still get your stuff out, you just can’t add more. If you’re running a small brand and cash gets tight for a quarter, that’s the difference between pausing a tool and losing a decade of institutional memory.
What Cross-Border Sellers Should Steal From This
You are almost certainly not going to migrate your product research into Harbor. But the design principles transfer directly to how you build your own internal stack.
Export clauses belong in your vendor scorecard. Before you sign another annual contract, ask one question: if we cancel tomorrow, what do we walk away with, in what format, and how long does it take? If the answer is “CSV of some fields” or “contact support,” you’ve found a single point of failure. Run that test on your 3PL, your returns platform, your review tool, your helpdesk. The vendors that pass are the ones worth multi-year commitments.
Own the retrieval layer, rent the intelligence. Harbor’s MCP server is the pattern. Your SKU data, supplier quotes, QC photos, and customer tickets should live somewhere queryable by whatever model is cheapest and best next quarter. If your product data only exists inside a dashboard, you can’t point an agent at it, and you’ll be re-entering it by hand every time the AI landscape shifts — which is roughly every eleven weeks.
Timestamp your own terms. The dated-note trick works internally too. When you agree a payment term with a supplier, a commission split with an affiliate, or a returns policy with a marketplace, write it into a dated document in a system you control, not just an email thread. Disputes are won by whoever has the artifact.
Where the math breaks
Harbor’s encryption trade-off is honest and worth understanding because it mirrors a mistake I see operators make constantly. When you encrypt a note, Harbor genuinely cannot read it — which means it cannot search inside it or OCR it either. Matthews’ guidance is to encrypt the tax returns and medical records and leave everything else fully searchable.
That’s a reasonable personal default. It’s a dangerous operational one. If you apply the same instinct to your business stack — encrypting or access-restricting “sensitive” data like supplier contracts or margin sheets — you break the exact search and automation layer that made the tool valuable. The right question isn’t “is this sensitive?” It’s “do I need a machine to read this, and do I trust the machine’s custodian?” Answer those two separately. Most operators collapse them and end up with a vault nobody can query.
Where My Judgment Says It Falls Short
No teams, on purpose. Matthews is upfront: no shared workspaces, no admin console, no enterprise tier coming. For a solo operator or a founder doing their own product research, fine. For a five-person DTC team where the sourcing manager, the ads buyer, and the CS lead all need the same supplier notes, this is a hard stop. You’d be paying for five personal accounts and manually syncing, which defeats the purpose. If your team is bigger than two, Harbor is a personal tool, not a company tool.
The incumbents aren’t standing still. Notion has the team collaboration Harbor refuses to build. Obsidian has local-first storage and a plugin ecosystem that already does most of what Harbor’s CLI and API promise, for free, if you’re willing to configure it. Apple Notes is free and native and good enough for 80% of personal capture. Harbor’s real differentiation is the combination — native apps without Electron, one-step Evernote import, OCR on handwriting, and a contractual price cap — and that combination is genuinely rare. But it’s a bundle advantage, not a moat. Any one of those competitors could ship the export guarantee next quarter.
The Evernote import is the whole growth engine, and it’s a shrinking pond. Harbor’s wedge is displaced Evernote users, and that population is real but finite and aging. The 54%-price-complaint stat is compelling today. In three years, the number of people who still have fifteen years of Evernote data to migrate will be a rounding error. Harbor needs a second act — probably the MCP/API angle, positioned as the retrieval layer for personal AI — and the launch copy doesn’t yet tell that story.
The three-year price lock has no enforcement teeth. I praised the dated note, and I stand by it as a trust signal. But let’s be clear-eyed: a note in your account is not a contract, and Cloudmanic Labs is a small shop. If Harbor gets acquired — the exact event that burned Matthews at Evernote — the new owner inherits the note and no legal obligation to honor it. The mechanism is clever. It is not a guarantee. Anyone treating it as one is repeating the mistake that created Harbor in the first place.
What I’d Watch / Test Next
This week, run an exit audit on your three most critical tools. Pick your product research platform, your email/SMS provider, and your 3PL. For each, actually attempt a full export of one month of data and time how long it takes. If you can’t complete it in an afternoon without contacting support, you’ve found a vendor you’re renting your business from rather than buying a service from.
Second, if you’re a solo operator or a two-person brand, give Harbor the free tier a real test — specifically the OCR on photographed supplier documents and handwritten QC notes. That’s the feature with the least competition at this price point, and it’s the one most likely to save you actual hours. If you’re on a team, skip it and instead steal the export-clause question for your next vendor negotiation.
Third, watch whether the MCP server gets real adoption. If Harbor becomes the default retrieval layer people point their Claude or ChatGPT at for personal and small-business data, the Evernote migration story becomes a footnote and the company has a durable second act. If it stays a note app with an API, it’s a well-built lifeboat with a shrinking harbor.
