Collections

What is a Collection?

A collection is a named group of vectors that share:

• A fixed dimension (number of floats per vector)
• A distance metric used for all searches

Every vector operation — upsert, search, delete — targets a specific collection.

┌─────────────────────────────────────┐
│ Collection: "product-embeddings"    │
│ Dimension: 384                      │
│ Metric: cosine                      │
├─────────────────────────────────────┤
│ doc-1 → [0.12, 0.34, ..., 0.89]    │
│ doc-2 → [0.55, 0.11, ..., 0.23]    │
│ doc-3 → [0.88, 0.67, ..., 0.01]    │
└─────────────────────────────────────┘

Creating a Collection

from vectordb_client import VectorDBClient

client = VectorDBClient(base_url="http://localhost:8000", api_key="your-key")

col = client.collections.create(
    name="product-embeddings",
    dim=384,
    distance_metric="cosine",
)

Parameters:

Parameter	Type	Required	Description
name	string	✅	Unique collection name. Letters, numbers, hyphens, underscores.
dim	integer	✅	Vector dimension. Must match your embedding model.
distance_metric	string	—	One of cosine, l2, or ip. Default: cosine.

Choosing a Dimension

Your dimension must match whatever embedding model produces the vectors.

Model	Dimension
all-MiniLM-L6-v2 (sentence-transformers)	384
text-embedding-ada-002 (OpenAI)	1536
text-embedding-3-small (OpenAI)	1536
text-embedding-3-large (OpenAI)	3072
BERT-base	768
nomic-embed-text (Ollama)	768

warning

The dimension is fixed at creation time and cannot be changed. If you need a different dimension, create a new collection and re-index.

Multiple Collections

Use separate collections to isolate different data types or models.

# Separate collections for different content types
client.collections.create("article-embeddings", dim=384, distance_metric="cosine")
client.collections.create("product-embeddings", dim=1536, distance_metric="cosine")
client.collections.create("user-profiles", dim=768, distance_metric="ip")

Listing and Getting Collections

# List all
collections = client.collections.list()
for col in collections:
    print(col.name, col.dim, col.vector_count)

# Get one
col = client.collections.get("product-embeddings")
print(col.vector_count)

Deleting a Collection

Deleting a collection removes all its vectors permanently.

client.collections.delete("old-collection")

danger

Collection deletion is irreversible. All vectors and metadata in the collection are permanently removed.