Text Chunker

About the Text Chunker

Long documents often do not fit into a single AI prompt, and retrieval systems (RAG) work better when text is cut into coherent pieces with a little overlap. The Text Chunker splits your paste into numbered chunks in the browser so you can copy them into ChatGPT, Claude, Gemini, notebooks, or your own pipeline — without uploading the file to UtilixTools.

This is a practical prep tool, not a hosted embedding service. Pair it with the Token Counter when you need a rough sense of size, and with AI Prompts when you want a clear instruction for each chunk.

How to use

1. Paste an article, transcript, notes, or export into the text box.
2. Choose how to measure size: characters, words, or ≈ tokens (a rough estimate, not an exact vendor tokenizer).
3. Set chunk size and optional overlap so neighboring chunks share a bit of context.
4. Prefer breaks on paragraphs or sentences when you want cleaner cuts; use hard cut only when you need exact sizes.
5. Click Split text, then copy one chunk or copy all with separators.

Why overlap helps

If you cut exactly at a size limit, a sentence or definition can land on the boundary and disappear from both sides of a retrieval or review pass. A modest overlap (for example 50–200 words, or a few hundred characters) keeps a little previous context at the start of the next chunk. Keep overlap smaller than the chunk size.

Choosing size

• Manual AI review — smaller chunks (roughly a few hundred to a couple thousand words) are easier to instruct and check.
• Prompt limits — leave room for your instruction, system message, and the model’s reply; do not fill the whole context window with source text alone.
• RAG-style prep — many teams use mid-sized chunks with overlap; exact settings depend on your embedder and search stack.
• Token estimate mode — useful for planning, but treat it as guidance. For billing-critical counts, verify with your provider’s tokenizer.

Tips for cleaner chunks

• Clean obvious boilerplate (menus, cookie banners, repeated footers) before splitting.
• Prefer paragraph breaks for articles and docs; sentence breaks for dense prose without blank lines.
• Label chunks when you paste into chat (“Chunk 2 of 7”) so the model knows the sequence.
• If a chunk is still too large for your model, lower the size and split again.
• For sensitive material, remember that copying into a third-party AI chat sends that text to the provider — this page only prepares the pieces locally.

Privacy

Chunking runs entirely in your browser with JavaScript. Your text is not uploaded to UtilixTools for processing. We do not store the document on our servers as part of this tool.

FAQ

Is the token mode exact for GPT-4o or Claude?
No. It is a practical estimate based on characters and words. Use it to plan splits; confirm exact counts in your own tooling if you need them.

Does this create embeddings?
No. It only splits text. You can feed the chunks into whatever embedding or chat workflow you use elsewhere.

What if overlap is larger than chunk size?
The tool will ask you to lower overlap. Overlap must stay smaller than the chunk size so progress does not stall.

Can I use this on a phone?
Yes. The form stacks on small screens; each chunk shows character/word/token estimates and its own copy button.