What is a lakebase?
Learn about a new generation of database built on object storage.
Lakebase Criteria
We define a lakebase as a database that meets the following criteria.
Operational Workload
It serves live, low-latency queries for apps and agents. That separates a lakebase from a lakehouse or a warehouse.
Compute-Storage Separation
Compute is stateless and ephemeral. It can start, stop, scale, or fail without losing data, because it owns no durable state.
Object storage source of truth
Durability lives in cloud object storage like S3, not on a disk attached to one machine. Local NVMe and RAM are caches, never the only copy.
Open-Format Storage
Data sits in open formats that other services can read directly. That keeps it portable, makes recovery fast, and removes the batch exports and data pipelines you would otherwise build.
Examples of lakebases
The clearest example is Lakebase Postgres. The category also spans vector, search, and time-series systems that meet the criteria. Each entry cites the vendor's own documentation.
Lakebase Postgres
Serverless Postgres built directly on cloud object storage.
turbopuffer
Search and retrieval with object storage as the sole source of truth.
Quickwit
Search that runs directly on object storage.
Zilliz Cloud and Milvus
A vector database with explicit storage and compute separation.
LanceDB
A vector database whose storage format is open and readable by other tools.
GreptimeDB
Time-series and observability data stored as Parquet on object storage.
SlateDB
An embedded storage engine that makes object storage the disk.
Side by side
How each example compares against the criteria.
| Database | Kind | Storage / compute split | Object storage source of truth | Open format | Storage backend |
|---|---|---|---|---|---|
| Lakebase Postgres | relational | Yes | Yes | Partial | Cloud object storage (S3) |
| turbopuffer | search | Yes | Yes | No | Object storage (S3, GCS, Cloudflare R2) |
| Quickwit | search | Yes | Yes | No | Object storage (S3, GCS, Azure, MinIO) |
| Zilliz Cloud and Milvus | vector | Yes | Yes | No | Object storage (S3, MinIO) for persistence |
| LanceDB | vector | Yes | Yes | Yes | Object storage (S3, GCS, Azure) |
| GreptimeDB | time-series | Yes | Yes | Yes | Object storage (S3, GCS, Azure) |
| SlateDB | key-value | Yes | Yes | No | Object storage (S3, GCS, Azure Blob, MinIO) |
Some databases separate storage and compute but are not lakebases. See what is not a lakebase: local-disk databases, analytics engines, and cloud-native databases on proprietary storage.