RAG Chunk Size Calculator
Connect corpus chunking, overlap cost and vector storage to a retrieval-plus-output context budget.
Enabling local controls…
Results
Review the inputs, then run to create a readable result and complete export. Editing cancels active work and clears stale results.
Formula, sources and limits
Independent document token counts are supplied assumptions. Retrieval upper bound = top-k × chunk size. Used = retrieval + system + question + output reserve. Deduplication, separators, chat wrappers and retrieval relevance require separate validation. Rates, limits, subscriptions and latency defaults are illustrative custom assumptions, not current vendor quotes or benchmarks. Check official sources and replace them. Each text field ≤32,768 UTF-16 units, total input ≤128 KiB, output ≤1 MiB; local job timeout 30s.
Methodology checked:
How to use this tool
Load the safe sample, review every labeled assumption, then run. Edit any field to clear old results. Copy or download the complete result. Processing stays in this tab.
When not to use it
Independent document token counts are supplied assumptions. Retrieval upper bound = top-k × chunk size. Used = retrieval + system + question + output reserve. Deduplication, separators, chat wrappers and retrieval relevance require separate validation. Rates, limits, subscriptions and latency defaults are illustrative custom assumptions, not current vendor quotes or benchmarks. Check official sources and replace them. Each text field ≤32,768 UTF-16 units, total input ≤128 KiB, output ≤1 MiB; local job timeout 30s.