Open-source Azure Cosmos DB alternatives

SurrealDB is a Rust-based multi-model database that brings several data models into one engine: key-value, document, graph, relational, time-series and geospatial data. Search and retrieval options include full-text, vector and hybrid, and it adds real-time and event-driven capabilities, queried through its SQL-like language, SurrealQL.

Because it supports end-user authentication, SurrealDB can also act as a backend-as-a-service, reducing the amount of server-side code an app needs. It ships as a single Rust binary that can run embedded inside an application, in the browser via WebAssembly, at the edge, as a self-hosted single node, or as a distributed cluster in the cloud. Typical workloads include data-intensive systems that need several data types, storage for AI agents, knowledge graphs, live applications such as recommenders and fraud detection, and embedded or edge deployments.

The repository license is listed as 'Other' on GitHub, so confirm the terms before commercial use. SurrealDB also offers cloud-hosted database service, and topics tag it as a serverless, realtime database for the web.

This repository holds the source of MongoDB, the document-oriented NoSQL database. It is written in C++ and contains the main server components: mongod, the database server, and mongos, the router used for sharded clusters. MongoDB stores flexible JSON-like documents rather than rows in fixed tables.

Most users install prebuilt packages rather than building from source. The README points to the community server download page, the mongodb-community-server Docker image and Homebrew taps, and also covers the MongoDB Shell, mongosh, and MongoDB Compass, a graphical client that can be installed with a packaged script. Client drivers exist for most programming languages, and RPM and Debian packages are produced by the build scripts. Documentation and a developer center help with learning.

The repository's license is listed as 'Other' on GitHub, so read the current license terms before building services on it. MongoDB can be run on your own infrastructure, and MongoDB Inc. also operates a managed cloud service. It is widely used as a general-purpose database for applications with evolving schemas.

ScyllaDB is a NoSQL database that implements the Apache Cassandra and Amazon DynamoDB APIs, so existing applications and drivers for those systems can talk to it. It is written in C++ on the Seastar framework and is aimed at real-time big data workloads that need high throughput and predictable latency.

The project describes a shared-nothing approach in which each CPU core manages its own slice of data, which it says increases throughput and storage capacity per machine and cuts hardware costs. Building the database is demanding: it needs a very recent C++23 compiler and many up-to-date libraries, so the project offers a frozen toolchain as a pre-configured Docker image that includes everything required. You then only need Docker or Podman to compile it.

The repository's license is listed as 'Other' on GitHub, so confirm the current terms before using it commercially. You can run ScyllaDB on your own servers or consume it as a managed cloud service from the company. It suits teams that have outgrown Cassandra's resource footprint or want a DynamoDB-compatible database they can run themselves.

ArangoDB is a multi-model database that stores documents, graphs and key-value data in a single engine. Instead of running separate systems for each data model, applications query connected data through one SQL-like language, with JavaScript extensions available for custom logic.

Every node in a graph is a JSON document, so existing document data can be imported and relationships can be queried across multiple levels of joins. ArangoSearch, an integrated search engine, adds text search and ranking. For scale and resilience the database can shard data horizontally across machines and replicate it to cluster nodes with automatic failover.

ArangoDB can be installed from a download or started in Docker, and runs on-premises or in any cloud. The Arango Managed Platform (AMP) creates deployments on major cloud providers and can be tried for free. A free Community Edition and a commercial Enterprise Edition without a dataset size limit are offered; the repository lists its license simply as Other, so check the terms before production use.

Apache Cassandra is an open-source, transactional, distributed NoSQL database. It is built to manage very large amounts of data across many machines, aiming for linear scalability and proven fault tolerance whether it runs on commodity servers or in the cloud, with no loss of performance. It is written in Java and developed under the Apache Software Foundation.

The project site presents Cassandra 5.0 along with documentation, a quickstart and community resources, and recent talks from the community cover topics such as Storage Attached Indexes, the Accord transaction protocol, mutation tracking, backups with Medusa, a sidecar process and analytics integrations. Cassandra is queried through CQL and can be run in clusters spanning racks, data centers or cloud regions.

Cassandra is licensed under Apache-2.0 and you run it yourself, either on your own servers or on cloud virtual machines or Kubernetes. It is typically chosen for write-heavy, always-on applications such as time-series data, messaging, IoT and large user-facing systems that must stay available when individual nodes fail.

Apache CouchDB is a database that stores JSON documents and is reached through an HTTP/JSON API. Its headline feature is multi-primary sync: the Couch Replication Protocol lets data flow between server clusters, mobile phones and web browsers, which supports offline-first applications that reconcile changes when connectivity returns.

It can run as a single-node database behind an application server or as a cluster, stores binary data alongside JSON, and offers a developer-friendly query language with optional MapReduce views. The protocol is implemented by a variety of projects and products across computing environments. CouchDB is written in Erlang, known for reliability, and the homepage presents the design as built for dependable operation.

Apache CouchDB is licensed under Apache-2.0 and runs on your own servers or with any cloud provider. Documentation, a technical overview and download packages are on the project site. It suits teams building apps that must keep working offline and sync later, or that want a simple, HTTP-native document store.

RavenDB is a NoSQL document database that is fully ACID and secure by default. Its makers, frustrated with the state of the database industry, aim to combine strong performance with ease of use so developers can concentrate on their application instead of managing data infrastructure. It is built on .NET.

Repository topics list full-text search, indexing, sharding, spatial queries, time series, vector search and IoT use, and the README links to documentation, articles, a book and a try-from-your-browser option. It describes itself as completely open source under AGPLv3 and mentions a free developer license, while GitHub reports the license as 'Other', so confirm the terms for production use.

RavenDB can be run on your own servers or consumed as a managed cloud service from the vendor, which also sells commercial licenses. It suits .NET teams that want a document store with transactions, indexing and search built in.

ArcadeDB is a multi-model database management system created by Luca Garulli, the founder of OrientDB, and described as a conceptual fork of it. It stores graph, document, key-value and time-series data in one engine, and supports ACID transactions, structured and unstructured data, a native graph engine that uses links rather than joins, full-text indexing and geospatial queries.

It can be queried in several ways: SQL, Cypher, Gremlin, HTTP and JSON, and it offers compatibility with MongoDB and Redis protocols. It also supports vector embeddings for similarity search, which its topics tie to vector database use. The engine is written in low-level Java 21 or newer to reduce garbage collection pressure, and it can run from a Raspberry Pi to multiple servers, embedded in an application, or in Docker and Kubernetes.

ArcadeDB is licensed under Apache-2.0 and is self-hosted on your own hardware or cloud. It reuses a modified version of OrientDB's SQL engine. It suits developers who want one database for connected data, documents and vectors rather than running separate graph and document stores.