Free tool • Runs in your browser
RAG text chunker
Split long text into overlapping chunks for embeddings and RAG pipelines, then copy them as JSON.
Quick answer
RAG systems search small passages instead of whole documents. Chunks of about 200 to 1,000 tokens with 10 to 20% overlap are a common starting point. This tool prefers paragraph and sentence boundaries so passages stay readable.
How to use it
- Paste your document.
- Set the chunk size and overlap.
- Review the chunks, then copy or download them as JSON.
Example
Chunk size: 500 characters Overlap: 50 characters Document: 1,200 characters
About 3 chunks, each with its index, text and length
Common errors and how to fix them
- Search results lack context
- Increase the chunk size or the overlap.
- Results are too broad
- Use smaller chunks so each one covers a single idea.
Frequently asked questions
What is chunking in RAG?
Chunking splits documents into smaller passages that can be embedded and searched. The system then gives the model only the most relevant passages.
How does RAG chunking work?
Text is divided into pieces of a set size, often with overlap. Each chunk is embedded and stored, then the closest chunks are retrieved for each question.
What chunk size should I use for RAG?
A common starting point is about 200 to 1,000 tokens with 10 to 20% overlap. Test with real questions, as the best size depends on your content.
Why use overlap between chunks?
Overlap keeps ideas that cross a boundary findable from either side, at the cost of some repeated text.
What is recursive chunking in RAG?
It tries larger separators such as paragraphs first, then smaller ones when a piece is still too long. This tool takes a similar approach by preferring natural boundaries.
What is semantic chunking in RAG?
It groups text by meaning, usually with embeddings, and splits where the topic changes. This tool does not do that, as it splits by size and text boundaries.
What is late chunking in RAG?
It embeds a whole document first and then creates chunk embeddings from it, which helps each chunk keep wider context. This tool only prepares text chunks.
How can I improve chunking in RAG?
Test different sizes and overlaps against real questions, keep headings with their text, and add metadata such as the source and section.
Can I chunk by tokens instead of characters?
Yes. Token mode uses a rough estimate of four characters per token, so check exact limits against your embedding model.
Is my document uploaded?
No. Chunking runs in your browser and your document is not sent to Quecko.
More AI tools
Need something built for your business?
Quecko designs and builds AI, Web3 and custom software products.