We built Ztract for the people who still retype documents by hand.

Why we started

We've watched accountants at bookkeeping firms work through stacks of invoices line by line — typing vendor names, dates, totals, and tax codes into spreadsheets one keystroke at a time. We've seen freight-forwarder operations staff retype the same fifteen fields from a commercial invoice into the customs portal, then again into their TMS, then again into the carrier's booking form. We've watched ops teams handle thousands of receipts a quarter the same way they would have handled them in 1995.

The technology to read those documents has been good enough for years. What was missing was a tool people could actually use without writing code, deploying a model, or buying a six-figure enterprise license.

So in May 2026, a few of us started Ztract. The premise is simple: a document-extraction service that a non-engineer can stand up in a single afternoon, that returns structured data you can drop straight into the tools you already use, and that doesn't charge you a monthly subscription you'll forget to cancel.

What Ztract does

You upload a document — PDF, Word, Excel, PowerPoint, image, or scan. You define what you want extracted — pick a template, describe it in plain English, or drop in a sample and let the engine infer a schema from it. You get back structured JSON, CSV, or Excel, with every extracted value anchored to its position on the source document so you can verify it in seconds. If something is off, you correct it in one click. No model training. No regex. No re-runs costing you more than the first run did.

It is, deliberately, a small product. We've focused it on the use cases we've watched cause the most pain — invoices, receipts, bank statements, contracts, IDs, resumes, lab reports, customs paperwork — and we'd rather do those well than do everything badly.

How we work

We're a small team, based in Hong Kong. We're fully bootstrapped — no outside investors, no growth-at-all-costs runway clock. That sounds like a footnote, but it shapes a lot of decisions. We can afford to keep prices simple and stable instead of raising them every quarter to hit a target. We can ship features only when they're actually ready instead of padding a release deck. We can say no to things — like training on customer data — that other companies say later to.

A word on your data

We do not train models on documents you upload. Not our engine, not the third-party LLMs we route through. The commercial APIs we use (OpenAI, Anthropic, Google) prohibit training on API-submitted data, and we rely on those commitments. When you delete a document, it's deleted — immediately from active storage and within 14 days from routine backups.

If you want the full picture of where your data goes and how long it stays, our Privacy Policy and Data Processing Agreement spell it out without hand-waving.