RAG Vector DB Cost Calculator

Estimate chunk count, embedding storage, vector index size, and monthly database cost for your RAG knowledge base.

Knowledge Base Configuration

Document Corpus

Number of Documents— Total documents to ingest

Avg Pages per Document— PDF/Word/HTML pages average

Tokens per Page— ~500 for dense text, ~250 for sparse

Chunking Strategy

Chunk Size (tokens)— Tokens per chunk (256–1024 typical)

Chunk Overlap: 20%— sliding window overlap

0% (no overlap)50%

Metadata Bytes per Chunk— Source URL, title, timestamps etc.

Embedding Model

1536 dims · float32 = 6,144 bytes/vector · $0.02/1M tokens

Vector Database & Query Load

Vector Database

Serverless pay-per-use. Queries billed as read units (vectors scanned × top-k).

Replication Factor— Copies of data (1 = no replication)

Queries per Day— Search/retrieval calls per day

Top-K per Query— Chunks returned per query

Configure your corpus and click Calculate

Results will appear here

Chunk size 256–512 tokens works well for most enterprise documents. Larger chunks reduce chunk count and storage but hurt precision retrieval.
Overlap at 10–20% improves recall for boundary-straddling concepts without ballooning storage significantly.
Smaller embedding dims (768d vs 3072d) can cut storage 4× with minimal quality loss for domain-specific corpora — consider fine-tuning before scaling up dims.
Self-hosted vector DBs (pgvector, Chroma) are cheapest at scale but require ops overhead. Use managed services (Pinecone, Qdrant Cloud) for fast time-to-production.
Re-ingestion cost is one-time. The main recurring cost is storage + query load. Cache frequent retrievals to reduce query billing.