cloudparts

Astra DB

DataStax’s serverless database built on Apache Cassandra, run on AWS, Google Cloud or Azure, with CQL tables, a JSON Data API and vector search; part of IBM since 2025.

What it is
Wide-column stores
Used for
Vector databases
Part of
Managed Apache Cassandra
Type
Commercial product
Made by
IBM (DataStax)
License
Proprietary
Facts checked
24 October 2026
Website

What it is

Astra DB Serverless runs Cassandra for you on AWS, Google Cloud or Azure. You create a database in a region and get keyspaces and tables as in Cassandra, reachable with CQL through the usual Cassandra drivers and a CQL shell. There is also a Data API, with clients for Python, TypeScript, Java, Go and C#, that treats a table as a collection of JSON documents with a flexible schema, and can sort results by similarity to a $vector. Indexes are Storage Attached Indexes (SAI), which allow filtering on non-key columns.

DataStax manages the cluster settings. Each region keeps three replicas across three availability zones, and a database can add regions, with data streamed from the primary region and arriving everywhere eventually. Writes at ANY, ONE and LOCAL_ONE are refused. Only the Unified Compaction Strategy is used, cassandra.yaml can’t be changed, and nodetool and JMX tools aren’t available. Limits include 200 tables and 100 indexes per database, 10 indexes per table, 75 columns per table and around 12,000 operations a second before rate limiting.

DataStax launched Astra as a hosted Cassandra service in May 2020 and re-engineered it in 2021 into a serverless service whose storage is separated from compute, billed by use. IBM announced it would buy DataStax in February 2025 and closed the deal in May; Astra DB is now sold within IBM watsonx, next to DataStax Enterprise, the self-managed Hyper-Converged Database (HCD) and Langflow.

Why teams pick it

  • Cassandra’s data model and drivers without running the cluster, on any of the three big clouds
  • One database for CQL tables, JSON document collections and vector search
  • Multi-region databases and a free plan for small projects
  • Zero Downtime Migration tooling for moving existing Cassandra clusters

Watch out for

  • Tight per-database limits: 200 tables, 100 indexes, about 10 Data API collections
  • No access to cluster settings, compaction strategy or nodetool; replication is fixed at three per region
  • Multi-region replication is eventually consistent and can lag for hours during failures
  • The free plan allows five databases, without private endpoints or multi-region replication

Good for

Specs

Compare with alternatives →
Data model
Cassandra tables, plus JSON document collections through the Data API
Query language
CQL, and the JSON Data API
Client protocol
Cassandra drivers (CQL); Data API clients for Python, TypeScript, Java, Go and C#
Secondary indexes
Storage Attached Indexes (SAI), up to 10 per table
Replication
Three replicas across three availability zones per region
Multi-region
Regions can be added; replication between them is eventually consistent
Storage engine
Cassandra, with the Unified Compaction Strategy only
Ways to run it
Managed only, on AWS, Google Cloud or Azure
Pricing model
Serverless, billed by use; free plan up to five databases

More of the Apache Cassandra family

What all wide-column stores have in common →Compare side by side →
Apache CassandraOpen sourceA masterless, wide-column database that spreads and replicates partitions around a ring of equal nodes, with consistency chosen per request.+1 Astra DBAmazon Keyspaces (for Apache Cassandra)AWSAlso good for scale to zero, pay per requestA serverless AWS database that speaks Cassandra’s CQL and drivers, with no nodes to run: you pay per read, write and gigabyte, and every write is stored three times across Availability Zones.ScyllaDBCommercialA C++ rewrite of Cassandra’s design with one shard per CPU core, speaking both CQL and the DynamoDB API.

See it in a real stack

No architecture uses Astra DB yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2010–2025

DataStax was founded in 2010, as Riptano, by Jonathan Ellis and Matt Pfeil to sell support for Apache Cassandra, and in 2011 it released DataStax Enterprise, a commercial Cassandra distribution. In May 2020 it launched Astra, a database service that ran Cassandra clusters in the cloud for customers.

In February 2021 DataStax announced a serverless version, generally available that March: it re-engineered the storage tier so it was separate from compute nodes, which let databases grow and shrink without a set size and bill only for what they used. Vector search followed in 2023, turning Astra DB into a vector store for AI applications alongside its Cassandra workloads.

IBM announced in February 2025 that it would acquire DataStax for its Astra DB, DataStax Enterprise and Langflow products, and closed the acquisition in May 2025. The products now sit in IBM’s watsonx portfolio, and the documentation carries IBM’s branding.

Why it’s built this way

Separate storage from compute
Classic Cassandra ties data to the nodes that serve it, so capacity is bought in nodes. Splitting the two let Astra scale each on its own and charge by use instead of by cluster size.
A JSON Data API on top of CQL tables
Fixed-schema tables suit structured rows, while collections with a dynamic schema suit semi-structured JSON and vectors. The Data API offers both over the same storage, so not every app has to design Cassandra tables.
Fixed replication of three across zones
DataStax sets the replication factor (three, with NetworkTopologyStrategy) and the compaction strategy itself and doesn’t let users change them, so every database is equally durable and maintenance can run region by region on a schedule.

Timeline

  1. 2010

    Riptano, later DataStax, is founded to support Cassandra

  2. 2020

    Astra launches as a cloud Cassandra service in May

  3. 2021

    Serverless Astra DB, with storage separated from compute, becomes generally available

  4. 2023

    Vector search comes to Astra DB

  5. 2025

    IBM announces in February and completes in May its acquisition of DataStax

Sources6 · facts checked 24 October 2026

Improve this page on GitHub