Smart Text Chunker & Slicer

Split massive documents into token-sized chunks while preserving paragraph, sentence, or word boundaries. Designed for AI context windows, RAG pipelines, prompt preparation, and large-document preprocessing.

📝 Your Text

Paste text or load a local document.

📂 Upload .txt / .md

Chars: 0 Words: 0 Est. tokens: 0 Paragraphs: 0

⚙️ Chunking Settings

Target chunk size

Cut preference

Overlap chunks Repeat the tail of the previous chunk for better RAG context continuity. Add


                                        [CHUNK n/N]

headers Adds an identifier when copying or downloading chunks.

Presets:

✂️

Paste or upload a document to see your chunks here.

📊 Summary

Total chunks: —
Largest chunk: —
Smallest chunk: —
Average chunk: —

Largest chunk vs target 0%

💡 How it works

Token estimates use approximately 4 characters per token.
Paragraph mode prefers paragraph boundaries and falls back to sentence boundaries when a paragraph is larger than the target.
Sentence mode attempts to split at sentence endings before moving to the next chunk.
Word mode keeps words intact wherever possible.
Hard mode is the only mode intentionally allowed to split anywhere.
Overlap repeats part of the previous chunk to improve context continuity.

Split text without losing context

Choose a target size, unit and boundary mode. Preview chunks before downloading text or ZIP. Paragraph and sentence modes keep those units intact when possible; a long unit can exceed the target.

Example: for a long article, start with paragraph boundaries and inspect oversized chunks. Word targets use actual whitespace-separated word counts. Hard mode may split words but preserves Unicode characters.

Token targets are approximate. Overlap repeats part of the previous chunk and increases size; headers also add text. Allow room for system instructions and the response. Verify each final chunk against your provider’s tokenizer.

Smart Text Chunker: common questions

How should I split text for RAG?

Start with paragraph or sentence boundaries so related ideas stay together. Choose a target size suitable for your retrieval task, inspect oversized chunks, and test whether retrieved passages contain enough evidence to answer your questions.

Are token-sized chunks exact?

Token targets are estimates. Boundaries, overlap, and added headers can increase final size. Leave space for other instructions and output, and check strict limits with the destination tokenizer.

What does chunk overlap do?

Overlap repeats text from the previous chunk to retain context near boundaries. It also increases the total text processed. Review whether the repeated content helps your retrieval task before increasing overlap.

Claupt tools are free to use without an account. Working text is processed in your browser. Site analytics and advertising follow your privacy choices.