Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/chroma-core/chroma/llms.txt

Use this file to discover all available pages before exploring further.

Collections are the fundamental organizational unit in Chroma. They are named groups of embeddings, documents, and metadata that you can query.

​
What is a Collection?

A collection is a container that holds:
  • Embeddings: Vector representations of your data
  • Documents: The original text or data
  • Metadata: Additional information about each record
  • IDs: Unique identifiers for each record
Think of a collection like a table in a traditional database, but optimized for vector similarity search.

​
Creating Collections

Create a new collection using create_collection(): 
import chromadb

client = chromadb.Client()

# Create a new collection
collection = client.create_collection(name="my_collection")

​
With Metadata

You can attach metadata to the collection itself: 
collection = client.create_collection(
    name="documents",
    metadata={"description": "Research papers", "type": "academic"}
)

​
With Custom Embedding Function

Specify a custom embedding function for the collection: 
from chromadb.utils.embedding_functions import OpenAIEmbeddingFunction

embedding_function = OpenAIEmbeddingFunction(api_key="your-api-key")

collection = client.create_collection(
    name="my_collection",
    embedding_function=embedding_function
)

​
Getting Collections

Retrieve an existing collection: 
# Get an existing collection
collection = client.get_collection(name="my_collection")

# Get or create (returns existing or creates new)
collection = client.get_or_create_collection(name="my_collection")

​
Listing Collections

List all collections in the database: 
# List all collections
collections = client.list_collections()

for collection in collections:
    print(collection.name)

# With pagination
collections = client.list_collections(limit=10, offset=0)

​
Deleting Collections

Delete a collection and all its data: 
client.delete_collection(name="my_collection")
Deleting a collection is permanent and cannot be undone. All embeddings, documents, and metadata in the collection will be lost.

​
Collection Operations

​
Count Records

Get the number of records in a collection: 
count = collection.count()
print(f"Collection has {count} records")

​
Peek

Quickly view the first few records: 
# Get first 10 records
results = collection.peek(limit=10)
print(results['ids'])
print(results['documents'])

​
Modify Collection

Update collection name or metadata: 
# Update collection metadata
collection.modify(metadata={"updated": "2024-01-01"})

# Rename collection
collection.modify(name="new_collection_name")

​
Collection Configuration

Collections can be configured with specific index and schema settings: 
from chromadb.api.collection_configuration import CreateCollectionConfiguration
from chromadb.api.types import HnswIndexConfig

configuration = CreateCollectionConfiguration(
    hnsw_index_config=HnswIndexConfig(
        space="cosine",  # or "l2", "ip"
        ef_construction=200,
        ef_search=100,
        m=16
    )
)

collection = client.create_collection(
    name="my_collection",
    configuration=configuration
)

​
Distance Metrics

Chroma supports three distance metrics (spaces):
  • cosine: Cosine similarity (default for most embedding functions)
  • l2: Euclidean (L2) distance
  • ip: Inner product

​
Schema Configuration

Define the structure of your collection data: 
from chromadb.api.types import Schema, StringValueType, FloatListValueType

schema = Schema(
    keys={
        "#document": StringValueType(),
        "#embedding": FloatListValueType(),
        "title": StringValueType(),
        "author": StringValueType(),
    }
)

collection = client.create_collection(
    name="my_collection",
    schema=schema
)

​
Indexing Status

Monitor the indexing progress of your collection: 
status = collection.get_indexing_status()

print(f"Indexed operations: {status.num_indexed_ops}")
print(f"Unindexed operations: {status.num_unindexed_ops}")
print(f"Total operations: {status.total_ops}")
print(f"Progress: {status.op_indexing_progress:.1%}")
This is useful for understanding when recent writes have been fully indexed and are available for search.

​
Best Practices

Choose collection names that clearly describe the data they contain:
# Good
collection = client.create_collection("product_descriptions")

# Avoid
collection = client.create_collection("data1")
Use collection metadata to track versioning and description:
collection = client.create_collection(
    name="documents",
    metadata={
        "version": "1.0",
        "created_at": "2024-01-01",
        "description": "Customer support documents"
    }
)
Different embedding models work best with different distance metrics:
  • Most OpenAI embeddings: use cosine
  • Some specialized embeddings: use l2 or ip
  • Check your embedding model’s documentation

​
Next Steps

Embeddings

Learn how Chroma handles vector embeddings

Metadata

Understand metadata and filtering

Querying

Query your collections with similarity search

Embedding Functions

Use embedding functions with collections

Build docs developers (and LLMs) love

Get started for freeTalk to us