Skip to content
Work

Gondnok

Works at dawn; you decide in the morning.

August 6, 2026·2 min readHeavy on jargon, but curious anyway? Read it in plain words, rewritten by AI.

In a five-hundred-product store nobody reviews the catalog by hand. At one end sit missing alt texts and expired sales; at the other, a cordless drill mispriced at 890 forints that took two orders yesterday. The owner has no time for the first and no eyes for the second.

Gondnok walks that round instead, at 5 AM. The daily touchpoint is a morning email with a decision list; its buttons work without logging in, because a "keep it / deactivate it" decision should never sit behind a login wall. If the owner ends up browsing the app, the product has failed.

gondnok.app
"Today" view: how much it handled this month, how much awaits a decision, and the open items with reasoning and two buttons
"Today" view: how much it handled this month, how much awaits a decision, and the open items with reasoning and two buttons
gondnok.app
The gondnok.app landing page on desktop and mobile
The gondnok.app landing page on desktop and mobile
The gondnok.app landing page on desktop and mobile

Write access is the stake

Gondnok writes: it fills in descriptions, prepares products, and changes statuses in a live store. One false positive can be enough for the owner to switch it off for good. Reliability is the product.

Rules decide what counts as a problem; the LLM triages the results and writes the explanation. The owner sets a separate trust level for each check: act on your own, ask me first, or leave it alone. The system saves the previous state before every change, so undo takes one click and has no time limit.

Daily to-dos, the per-check trust ladder, and the log with undoable entries on a phone
Daily to-dos, the per-check trust ladder, and the log with undoable entries on a phone
Daily to-dos, the per-check trust ladder, and the log with undoable entries on a phone
Daily to-dos, the per-check trust ladder, and the log with undoable entries on a phone

Three layers, one email

Catalog hygiene keeps the store tidy: missing titles, descriptions, alt texts, wrong prices, expired sales. The guard protects revenue: suspiciously cheap orders immediately, stock-out forecasts, stuck inventory, competitor price checks. New-product preparation works ahead: it fills in what the data supports and drafts or lists the rest. All three end in the same single email, and the report speaks in forints, not error counts: how much turnover a suspicious price touches, how much revenue the out-of-stock bestsellers are losing.

The boundary is the same everywhere: a suggestion stays a suggestion until the owner approves it. Gondnok publishes nothing on its own.

New-product preparation: tags for the missing fields, with options to skip, search the web, or use an AI suggestion
New-product preparation: tags for the missing fields, with options to skip, search the web, or use an AI suggestion

Paper, ink, pine green

The usual AI gradient would mislead here. The interface uses paper, printing ink, and a single pine green. Its landing page centers on a daily statement printing itself out, echoing what the product does every morning.

Before the pilot

Write tests and undo are proven on the test store. The entry point is a free, read-only audit, then a fourteen-day trial with no card. Success has one metric: is it still switched on two weeks later.

Tell me what you're working on.

Happy to talk about a product, a collaboration, or a role. A few sentences is enough to start.