ashwin.system60 fps

~/tools / rag-planner

RAG capacity planner

Every RAG system starts with the same arithmetic: how many chunks, how many vectors, how much RAM, what will embedding cost. This planner runs that math live — documents, average length, chunk size, overlap, dimensions — and tells you what your corpus really weighs. It's the exact sizing pass I ran before indexing 36,000 documents.

TOOL 02

RAG capacity planner

the sizing math I ran before indexing 36K documents
9
chunks per document
324,000
total vectors
113.4 MTok
tokens to embed
$2.27
one-time embed cost
2.79 GB
index RAM (HNSW +40%)
1.99 GB
raw vector storage
That's 1.0× the corpus I run at Jagriq (~329K units). Overlap above ~20% mostly buys you bill, not recall.
more

tools.siblings

careertime INFO boot: system online  0%