Size first, evaluate next
Estimate RAG chunks and vector storage
Use the estimate to plan an evaluation, not as a universal chunk-size recommendation.
Provider-neutral local estimate
RAG Chunk & Storage Estimator
Estimate chunk count, overlap work, vector bytes, metadata, replication, and retrieval context. The result is a sizing aid; evaluate retrieval quality on your own documents.
Initial sizing estimate
- Chunks
- 1,471
- Step after overlap
- 680 tokens
- Embedding tokens processed
- 1,176,400
- Raw vectors
- 9.04 MB
- Metadata
- 735,500 B
- Replicated total
- 19.55 MB
- Context per query
- 4,800 tokens
- Optional storage estimate
- Not entered
Chunk-size sensitivity (overlap unchanged)
| Scenario | Chunk size | Chunks | Embedded tokens | Query context |
|---|---|---|---|---|
| −25% | 600 | 2,084 | 1,249,960 | 3,600 |
| Current | 800 | 1,471 | 1,176,400 | 4,800 |
| +25% | 1,000 | 1,137 | 1,136,320 | 6,000 |