Frequently asked questions

You already pay for ChatGPT or Claude - but the PDFs you feed them often trip the AI on layout before it ever reads your content. LessTokens prepares PDF, Word, Excel and more into clean, AI-ready Markdown, and openly tells you what it could not read. A prepared document makes the AI subscription you already pay for worth more.

PDF, Word, Excel, PowerPoint, HTML, CSV, TXT, Markdown - plus scans and screenshots via free OCR. Excel converts what you see: 12.5% stays 12.5%, never 0.125; currency symbols, thousands separators and negative parentheses are kept.

Files are deleted automatically within 60 minutes and are never used to train any AI model. Conversion runs entirely on our own servers and never passes through an external paid AI service - that is architecture, not marketing: even the OCR engine is self-hosted, so your document never has to leave for a third party.

Currently completely free, OCR included. Paid plans are being designed around added value for heavy and professional use - not around locking up what is free today.

You can - many people do, right up until a number silently shifts, a table scrambles, or a two-column report gets read straight across. The trouble is that an AI given bad input will not warn you; it answers confidently anyway. LessTokens fixes the structure first and puts conversion risks in plain sight, so you know what the AI received before it answers.

Our rule is honesty first: if a conversion fails, we say so instead of handing you an empty file dressed up as success. Unreadable characters, suspect numbers and unmeasured items are listed right next to the result. You do not have to believe we are perfect - you only need to know where to double-check the source, and that is exactly what we mark for you.

The honest answer: large documents measure around ten percent in our tests, and small ones can even grow slightly, because structure itself costs characters. But the point is usually fit, not cost - the thirty-odd thousand tokens saved on a 320k-token document is often the difference between read in full and cut off. Convert one document and you will see your own number.

Taiwan's financial reports, contracts and technical documents are rarely pure Chinese or pure English - and that is exactly our home ground. Mixed scripts, full- and half-width characters, NT$ and percentages are what we test against every day.

Yes - OCR is completely free (self-hosted, up to 20 pages per file). And we attach confidence warnings: passages the OCR was unsure about and numbers with low confidence are flagged, instead of pretending every character came out right.

Copy it straight into ChatGPT, Claude, Gemini or NotebookLM - or copy it as a hand-off prompt that tells the receiving AI to respect the conversion warnings before answering. For documents where figures matter, a dedicated mode reminds you to verify the source before copying.

Write to support@lesstokens.ai - the founding team reads every message. This product ships updates weekly; the release notes in the footer are the proof.