Free Custom GPT Knowledge Base Builder

How it works

1. 1

   Drop in a document

   PDF, Word, CSV, Markdown, HTML, or plain text. No account, no card, no email.

2. 2

   We clean and chunk it

   Page furniture and broken formatting come out. What is left gets split into retrieval-sized chunks with overlap.

3. 3

   Download and upload

   Plain .txt files with a token count on each one. Drop them straight into a custom GPT, a Claude Project, or your own RAG stack.

Why a cleaned, chunked file beats a raw export

When a custom GPT gives a confident wrong answer, the model usually is not the problem — the input is. Export a PDF and the headers, footers, page numbers, and broken column layouts all come through as text. The retriever cannot tell the difference between a page footer and a fact, so it matches on the wrong passage and the model answers from it.

Size matters just as much. Upload one 200-page document and every question gets matched against the same enormous blob. Split that into a few hundred focused chunks, give each one a little overlap with its neighbours so an idea that spans a boundary stays findable from either side, and retrieval suddenly has something to aim at.

That is the whole job this tool does — and it is the step most PDF converters skip. They hand you one large Markdown file and leave the splitting to you. Here is how to think about chunk size and overlap if you want to tune it yourself.

What you get back

• ✓Plain .txt files — no proprietary format, nothing to convert again
• ✓An approximate token count on every chunk, so you know it fits before you upload
• ✓Source attribution on each chunk — you can always trace a passage back to its file
• ✓Overlap between neighbouring chunks so boundary-spanning ideas stay retrievable

Frequently asked questions

Is this really free?

Yes. One file up to 5MB per conversion, no account and no credit card. Paid plans exist for batch processing, larger files, and custom chunk settings, but nothing on this page is gated.

Do you store my documents?

No. Text-based formats are processed entirely in your browser and never leave your device. PDFs are sent to our server only to extract their text layer, and the file is discarded as soon as the text is returned — it is never written to disk or a database.

What file formats can I use?

PDF, Word (DOCX), plain text, CSV, Markdown, and HTML. PDFs need a real text layer — if the page is a scanned image, there is no text to extract and you will need to run OCR on it first.

What do the output files look like?

You get plain .txt chunks. Each chunk carries a header with its number, the source file it came from, its key terms, and an approximate token count, so you know it fits before you upload it anywhere.

Why does chunking matter for a custom GPT?

Retrieval works by matching your question against pieces of your documents. One giant file gives the retriever nothing to aim at, so it returns loosely related passages. Chunks sized to a few hundred tokens, with a little overlap so ideas that span a boundary stay findable, are what make a custom GPT answer from the right passage.

How is this different from a PDF-to-Markdown converter?

Converters hand you one large file and stop. That is extraction, not preparation — you still have to split it yourself. This tool finishes the job: it cleans the text, splits it into retrieval-sized chunks with overlap, and labels each one with its token count.

Related guides