Chunking & Retrieval Playground
Paste a document, move the sliders, and watch how chunk size and overlap reshape the passages your retriever will index — and why that changes answer quality.
Document & settings
The chunks
Ochre highlight = text carried over from the previous chunk (the overlap).
Illustrative. Real RAG pipelines may chunk by tokens, sentences or document structure, and add metadata. This shows the core size/overlap trade-off; tune against your own documents and queries.
Good to know
Why does chunking matter?
Retrieval works on chunks, not whole documents. Chunk size and overlap decide whether the model gets focused, complete context — or noisy fragments. It’s one of the biggest levers on RAG answer quality.
Characters or tokens?
This tool chunks by characters for an exact, transparent split. As a rule of thumb, tokens ≈ characters ÷ 4 for English, so a 600-character chunk is roughly 150 tokens.
What are good starting settings?
Many setups start around 500–1,000 characters with 10–20% overlap, then tune against real queries. There’s no universal best — it depends on your documents and how people ask.
Want this guaranteed in your own infrastructure?
Get a copy of these results by email and see RAGSuite running on a setup like yours — citation-backed, self-hosted, EU-ready.
Thanks — we’ll be in touch within one business day (DE/EN).
Chunking is one lever. Measuring is the other.
RAGSuite pairs retrieval with Compare Models and deep query tracing, so you can see which settings actually improve answers — not just guess.