Quick start
-
Set params
Same chunk size and overlap as the chunk splitter.
-
Read metrics
Watch avg chunk length and overlap waste.
RAG Chunk Analyzer
Enter a document and chunk params to see count, char distribution, and overlap waste. Tune RAG indexing without guessing chunk size.
Privacy: processed locally, never uploaded.
↓ Paste in the input area below to see results instantly
Analyze chunk stats: count, avg/min/max chars, overlap waste. Tune alongside the chunk splitter.
Chunk size
Overlap
Document text
Chunks
1
Avg chars
138
Min
138
Max
138
Overlap waste
0%
Empty chunks
0
Notes
Reading stats
Overlap waste is redundant chars from overlapping windows. Too many tiny chunks may hurt retrieval quality.
Enter a document and chunk params to see count, char distribution, and overlap waste. Tune RAG indexing without guessing chunk size.
With chunk splitter
This outputs stats only; the chunk splitter shows chunk text. Analyze first, then split.
Features and use cases
Preview RAG chunk sizes and overlaps by characters or estimated tokens to tune retrieval strategy.
Use before building KBs to compare chunk_size impact on recall and tune vector pipeline configs.
Typical Workflow
When using the RAG Chunk Analyzer, first paste your document text or upload a file, then set the chunk size and overlap. The tool instantly calculates the number of chunks and displays character distribution via histogram. By adjusting sliders, you can observe wasted overlap characters under different parameters to find the optimal chunking strategy balancing retrieval accuracy and storage efficiency.
In practice, start with small chunks (200-500 chars) for technical docs and larger chunks (800-1500 chars) for articles. Reduce overlap when the waste rate exceeds 15%. After exporting results, apply these parameters to your vector database indexing pipeline to avoid repeated trial-and-error.
Examples
Example
Input
500 chars, 50 overlap
Output
6 stat metrics
FAQ
How is waste calculated?
(total chunk chars − raw length) / total chunk chars; reflects overlap redundancy.
Why does my overlap waste rate spike suddenly?
When chunk size and overlap are integer multiples (e.g. 1000 chars + 500 overlap), text gets redundantly calculated. Prevent this by using prime numbers (e.g. 997 chars) or fine-tuning overlap to 10%-20% of chunk size.