Vector Databases

Chroma vs turbopuffer

Compare Chroma and turbopuffer side-by-side. See feature differences and pricing in the Vector Databases category.

by Chroma

No reviews yet

Pricing: Open-source engine free to self-host. Chroma Cloud: Starter $0 a month plus usage, Team $250 a month plus usage, Enterprise by quote; monthly terms, no annual plan.

Chroma is an open-source vector database for AI retrieval, licensed Apache 2.0 to self-host or run as Chroma Cloud, a serverless service billed by usage.

8 /18 features Full profile

turbopuffer

by turbopuffer

No reviews yet

Pricing: Usage billed against a monthly minimum: Launch $16, Scale $256, Enterprise $4,096 or more plus a 35% usage premium. Storage up to $0.33/GB, writes $2/GB, queries $1/PB.

turbopuffer is a serverless vector and full-text search database on object storage, for teams running AI search over large multi-tenant data. Billed by usage.

13 /18 features Full profile

Feature Comparison

Vector Databases features side-by-side

Feature	Chroma 8/18 features	turbopuffer 13/18 features
Approximate nearest-neighbor index An HNSW, IVF or similar index serves top-k similarity queries in milliseconds over millions of vectors.	Yes	Yes
Filtered vector search on metadata Structured filters on payload fields applied inside the similarity search, not after it.	Yes	Yes
Hybrid keyword and vector search Sparse or keyword signals combined with dense embeddings in one ranked query.	Yes	Yes
Choice of similarity metrics Cosine, dot product and Euclidean distance selectable per collection or index.	Yes	No
Quantized index compression Scalar, product or binary quantization shrinks index memory with a tunable recall trade-off.	No	No
Multi-tenant namespaces per index Tenant data isolated inside one index so a per-customer query never scans other tenants.	No	Yes
Integrated embedding models Text or images embedded by a hosted model at write and query time, with no separate pipeline.	Yes	Yes
Built-in reranking stage A second-stage reranker model reorders the top candidates before results are returned.	No	Yes
Multiple named vectors per record Several vectors per record, such as text and image embeddings, searchable together.	No	Yes
Live upserts visible in seconds New or updated vectors become searchable within seconds while the index stays online.	No	Yes
Sharded vector collections Collections split across nodes for datasets beyond one machine's memory.	No	Yes
Replicated vector shards Each shard is copied to another node so queries keep answering through a node loss.	No	Yes
Disk-backed index beyond RAM Indexes served from SSD rather than memory so billions of vectors fit on modest hardware.	No	Yes
Collection snapshots and restore Point-in-time snapshots of a collection that can be restored or copied to another cluster.	Yes	Yes
Role-based keys per collection API keys or roles scoped to specific collections and to read or write operations.	No	No
Bulk import from Parquet files Vectors and metadata loaded in bulk from Parquet or JSON files in cloud storage.	No	No
Serverless usage-based pricing Billing by reads, writes and stored vectors, with no cluster to size in advance.	Yes	Yes
Open source self-hosted engine The engine's source is public and can be self-hosted without a commercial license.	Yes	No
Overall Rating	No reviews yet	No reviews yet

Chroma 8/18 features

Approximate nearest-neighbor index: Yes
Filtered vector search on metadata: Yes
Hybrid keyword and vector search: Yes
Choice of similarity metrics: Yes
Quantized index compression: No
Multi-tenant namespaces per index: No
Integrated embedding models: Yes
Built-in reranking stage: No
Multiple named vectors per record: No
Live upserts visible in seconds: No
Sharded vector collections: No
Replicated vector shards: No
Disk-backed index beyond RAM: No
Collection snapshots and restore: Yes
Role-based keys per collection: No
Bulk import from Parquet files: No
Serverless usage-based pricing: Yes
Open source self-hosted engine: Yes

No reviews yet

turbopuffer 13/18 features

Approximate nearest-neighbor index: Yes
Filtered vector search on metadata: Yes
Hybrid keyword and vector search: Yes
Choice of similarity metrics: No
Quantized index compression: No
Multi-tenant namespaces per index: Yes
Integrated embedding models: Yes
Built-in reranking stage: Yes
Multiple named vectors per record: Yes
Live upserts visible in seconds: Yes
Sharded vector collections: Yes
Replicated vector shards: Yes
Disk-backed index beyond RAM: Yes
Collection snapshots and restore: Yes
Role-based keys per collection: No
Bulk import from Parquet files: No
Serverless usage-based pricing: Yes
Open source self-hosted engine: No

No reviews yet

Pricing

Chroma

Open-source engine free to self-host. Chroma Cloud: Starter $0 a month plus usage, Team $250 a month plus usage, Enterprise by quote; monthly terms, no annual plan.

See all plans

turbopuffer

Usage billed against a monthly minimum: Launch $16, Scale $256, Enterprise $4,096 or more plus a 35% usage premium. Storage up to $0.33/GB, writes $2/GB, queries $1/PB.

See all plans

Explore Each Product

Chroma Read reviews, features & pricing

turbopuffer Read reviews, features & pricing

Looking for more options?

Browse all Vector Databases