# RAG Embedding Dimensions and Index Compatibility
Status: public
Confidence: medium (0.725) (verified)
Last verified: 2026-06-02
Generation: ai_structured


## TL;DR

Embedding dimensions and index configuration must match, or RAG ingestion and retrieval can fail or silently mix incompatible vectors.

## Core Explanation

Vector stores are configured around assumptions about vector dimensionality and distance metric. If an agent switches embedding models without reindexing or migration, new vectors may not fit the existing index or may be compared under the wrong metric.

Agents should inspect index dimension, vector field schema, embedding model version, metric, normalization, and whether a backfill completed. Reindexing is often part of a model upgrade, not an optional cleanup step.

## Source-Mapped Facts

- Pinecone documentation describes the create index API as taking a dimension value for the vectors stored in the index. ([source](https://docs.pinecone.io/reference/create_index/))
- Qdrant documentation describes vector size as the dimensionality of vectors for a collection. ([source](https://qdrant.tech/documentation/manage-data/vectors/))
- Weaviate documentation describes vector indexes as enabling similarity search over stored vectors. ([source](https://docs.weaviate.io/weaviate/concepts/vector-index))

## Further Reading

- [Pinecone Create Index](https://docs.pinecone.io/reference/create_index/)
- [Qdrant Vectors](https://qdrant.tech/documentation/manage-data/vectors/)
- [Weaviate Vector Index](https://docs.weaviate.io/weaviate/concepts/vector-index)