Knowledge-base inputs
Free RAG planning tool
Free RAG Chunk Size Calculator
A RAG chunk size calculator estimates how documents will be split for retrieval-augmented generation. Plan chunk count, overlap, embedding volume, retrieval depth, and context-window fit before you ingest a knowledge base.
Chunk + overlap mathEmbedding volumeContext fitPrivate browser calculation
Plan my chunks Chunking guidance
How to choose a RAG chunk size
1. Start with structure
Prefer headings, paragraphs, tables, or code blocks as boundaries. Token targets work best when semantic boundaries are preserved.
2. Test 400-1,000 tokens
Smaller chunks favor retrieval precision. Larger chunks carry more context. Compare answer quality on a representative evaluation set.
3. Budget the whole prompt
Leave room for instructions, the question, metadata, chat history, citations, and output instead of filling the entire context with retrieved text.
Frequently asked questions
What is a RAG chunk size calculator?
A RAG chunk size calculator estimates how a document collection will be split for retrieval-augmented generation. It shows chunk count, overlap, embedding volume, retrieval context, and whether the selected chunks fit a model's context window.
What is a good chunk size for RAG?
A practical starting point is 400-1,000 tokens. Smaller chunks can improve precision, while larger chunks preserve more context. The best size depends on document structure, query style, embedding model, and reranking strategy.
How much overlap should RAG chunks use?
Start with 10-20% overlap. That usually preserves context across boundaries without duplicating too much embedding and retrieval volume. Structured documents may need less overlap; narrative text may benefit from more.
How does retrieval depth affect context usage?
Retrieval depth is the number of chunks placed into the model prompt. Approximate retrieval tokens equal chunk size multiplied by depth, before metadata and separators. A higher depth adds evidence but consumes more context.
Does this calculator use my documents?
No. The calculator uses only the numeric assumptions you enter and runs entirely in your browser. It does not upload or inspect document content.
Related AI tools
LLM Token Counter
Count tokens and compare per-request model costs.
Open toolLLM VRAM Calculator
Estimate model weights, KV cache, and GPU memory.
Open toolAI API Pricing Calculator
Compare token pricing across major AI APIs.
Open toolGet practical positive AI news, implementation ideas, and tools from AI Wins.
Explore AI Wins