Developer tools

RAG size estimator

Estimate overlapping chunks and raw vector storage.

Processed locally

Estimate for one continuous corpus. Excludes document boundaries, index metadata, replicas and padding.

Your answer will appear here.

Help and method

Stride = chunk − overlap. Include a final partial chunk; repeated tokens = overlap × (chunks − 1). Raw bytes = chunks × dimensions × bytes per dimension. Whole counts only; overlap must be smaller than chunk. Corpus ≤10¹²; chunk ≤10⁷; dimensions ≤10⁶; bytes: 1–16.

Token counting and scopeMethods and reliability

Worked example

An example you can try or adapt to your own values.

Corpus tokens
1000
Chunk tokens
400
Overlap tokens
100
Vector dimensions
3
Bytes per dimension
4

Result

1,000 corpus, 400 chunk, 100 overlap, 3 dimensions, 4 bytes → 3 chunks, 1,200 embedding tokens, 36 bytes.