en

Prepare Files for AI

Turn PDF, text, Markdown, CSV and Word files into one local AI context pack with source references, chunks and a privacy check, with no upload.

Running locally on your device ...

Output The target format of the output file.
  • Markdown (.md)
  • Text (.txt)
  • ZIP with separate files
Chunk size (tokens, estimated)
Context window of your model (tokens, 0 = do not check)
Reserve for question and answer (tokens)
Personal data
  • Mask
  • Report only
  • Do not check
Replace file names with neutral names
Text recognition (OCR) for pages without a text layer

Running locally on your device ...

0%

Is my file uploaded?

No. Everything runs in your browser - your file never leaves your device. How this is verifiable

No upload100% local
Your content stays with youno third-party access
Hosted in GermanyGlobal Content Delivery
Independently auditedTLS A+ · HTTP headers A+

Anyone who hands documents to an AI assistant usually pastes text together from several files and loses track of where each passage came from. This tool reads PDF files with a text layer, plain text, Markdown, CSV, DOCX and ODT with the extractors this site already uses and builds one context pack from them: a source manifest at the top with file name, type, size, SHA-256 checksum, extent and status, then a table of contents built from the headings, then the text sections. Every section names its source and its page or line range, so you can trace any answer of the AI back to the passage it came from.

The sections (chunks) are formed against a token limit you set. Counting uses a clearly labelled estimate by writing system, not the tokenizer of a particular model. If you enter the context window of your model and a reserve for question and answer, the report checks whether the pack fits or should be sent section by section; model values are deliberately not pre-filled. Headings from Markdown and Word open a section, CSV sections repeat the header row, and over-long lines are split at character boundaries, including Chinese, Japanese or emoji.

Before anything is output, a check looks for e-mail addresses, phone numbers, IBANs and card numbers (with check digit) as well as API keys. By default they are replaced by placeholders such as [EMAIL_1]; the report names only the kind and the location, never the value. Metadata such as the author or the editing history never enters the pack, and file names can be replaced by neutral names. Scanned PDFs without a text layer are reported; with text recognition (OCR) switched on, the tool reads those pages in the language you choose. Copy-protected or encrypted files are not bypassed. Everything runs in your browser, offline too after the first load.

Many AI chats cap file size, file count and the context window in tokens - an oversized file gets rejected or only partially read. Prepare Files for AI prepares your file locally in your browser before you upload anything at all.

Specifications

Specifications
Input formatsPDF, TXT, TEXT, LOG, MD, MARKDOWN, CSV, TSV, DOCX, ODT
Auto-conversionPNG, JPG, WebP, GIF, BMP, TIFF, ICO, SVG, HEIC, MP3, WAV, ZIP, XLSX, SQLite, DBF, VCF, ICS, 7Z, DXF, MIDI, GeoJSON, DDS, HDR, PCX, WBMP
Output formatMD
Batch processingYes
ProcessingLocally in your browser (JavaScript)
File uploadNone

In 3 steps

  1. Choose or drop your files, several at once if you like.
  2. Choose the output format and chunk size, enter the context window of your model if you like, and for scanned PDFs switch on text recognition with the document language.
  3. Check the report, then copy the pack or download it as Markdown, text or ZIP.

Limitations: Token counts are estimates and can differ from the tokenizer of your model. The existing text layer is read; tables, columns and formulas are not rebuilt perfectly, and text recognition can make mistakes. The personal-data check is a heuristic and does not find names or postal addresses. Archives (ZIP, GZIP, 7z, RAR, TAR) and spreadsheet files (XLS, XLSX, ODS) are not opened, PowerPoint is not read. Limits: 50 files, 100 MB per file, 300 MB in total, 20 million characters of text, Office files at most 64 MB unpacked. Pack and report use short English labels that this page explains.

FAQ

Are my files uploaded?

No. Reading, text recognition, the personal-data check and building the pack all run in your browser. Nothing is stored and no connection to any AI service is made; what you later give to an AI is your own decision.

How exact is the token count?

It is an estimate by writing system with a range, not the tokenizer of a particular model. For the fit check the report also uses the upper estimate, so that a pack counts as too large rather than too small when in doubt.

What happens to e-mail addresses and IBANs in my files?

In Mask mode the tool replaces every detected value with a placeholder such as [EMAIL_1] or [IBAN_1]; equal values get the same placeholder. In Report only mode the text stays unchanged and the report lists the locations without the values.

Why does my PDF show no_text_layer?

The PDF is scanned or made of images only and has no text layer. Switch on text recognition and choose the language of the document; the report then lists the pages read by OCR.

Why isn't uploading a file straight into an AI chat always enough?

AI chats cap file size, file count and the context window in tokens; an oversized file gets rejected or only partially read. Prepare Files for AI splits or prepares your file beforehand, locally - with no upload.

Related tools

PDF to Markdown · DOCX to Markdown · AI Token Calculator · Compare prompts · Pseudonymizer before AI chat · Translate text (German/English)