DingoDB is an Apache-2.0 licensed, Java-based distributed multi-modal vector database from DataCanvas that unifies relational and vector semantics behind a MySQL-compatible SQL interface, built for teams that need hybrid scalar-and-vector queries over structured and unstructured data at high concurrency and ultra-low latency.

What it is

DingoDB is an open-source distributed multi-modal vector database independently designed and developed by DataCanvas. It integrates real-time strong consistency, relational semantics, and vector semantics into one platform, and it treats Table and Vector as first-class citizen data models rather than forcing embeddings into a side channel. Access is offered through SQL, SDK, and API interfaces, with seamless compatibility with the MySQL protocol, and the system is built around horizontal scalability, elastic scaling, and enterprise-grade high availability. The repository lives in the AI and Machine Learning ecosystem under Machine Learning Infrastructure, and its topic list places it across embedding-search, embedding-store, hybrid-search, real-time-semantic-search, vector-database, unified-sql, structured-data, and unstructured-data.

The concrete problem it solves is the split that appears when a relational database holds scalar data while a separate vector index holds embeddings: two systems, two query languages, and a synchronisation path between them. DingoDB replaces that arrangement with scalar-vector hybrid retrieval inside a single engine, supporting both traditional database index types and multiple vector index types, along with fusion of scalars and vectors and distributed transaction processing. Because DingoDB is MySQL-protocol compatible, existing MySQL clients can reach it without adopting a new dialect, and cold-hot tiered retrieval lets disk-based vector search carry massive datasets while memory consumption stays under control.

Key capabilities

• Unified SQL access over structured and unstructured data, with MySQL protocol compatibility plus SDK and API access modes.
• Table and Vector as first-class citizen data models in the same system rather than one derived from the other.
• Scalar-vector hybrid retrieval across traditional database index types and multiple vector index types, including fusion of scalars and vectors with distributed transaction processing.
• Built-in real-time index optimization: scalar and vector indexes are built in real time, with unconscious background automatic index optimization and no delay during data retrieval.
• Fully automatic elastic data sharding with dynamically configurable shard size, automatic splitting, and automatic merging.
• Cold-hot tiered retrieval for massive datasets, using disk-based vector search to minimize memory consumption and dynamic switching between indexes based on data scale.
• Built-in data high availability that requires no external components to deploy, reducing deployment and operation and maintenance cost.

Who uses it and how

• Teams running semantic search and embedding storage alongside ordinary relational tables, who want one query surface instead of a database plus a separate vector index.
• Operations groups that deploy a single node with Docker for evaluation, then move to a multi-node cluster provisioned with Ansible.
• Enterprises with high-concurrency, low-latency serving requirements that also need strong consistency and distributed transactions rather than eventual consistency alone.
• Applications with large embedding collections where memory is the binding constraint, using disk-based vector search and tiered index switching.
• Multi-language application stacks that already speak MySQL, since compatibility with the MySQL protocol lets existing drivers and tooling connect without a bespoke client.

Getting started

Full documentation is published at dingodb.readthedocs.io, and installation is documented for two paths: deploying a single node using Docker, or deploying on a cluster by Ansible. The Dingo-Deploy project handles deployment of compute nodes and storage nodes, while Dingo-Store provides the strongly consistent distributed storage layer based on the Raft protocol.

How it compares

No paid or competing product list is provided in this registry entry, and the facts name no similar alternative to compare against, so DingoDB stands alone in this registry on that axis. The related repositories named in the README, Dingo-Store and Dingo-Deploy, are its own storage and deployment components rather than substitutes for it.

When to use it — and when not to

A self-hoster must operate both compute nodes and storage nodes, with Dingo-Store providing the Raft-based distributed storage underneath, so the operational burden is a real cluster rather than an embedded library. Teams wanting a managed hosted service, a single-binary embedded vector index, or a stack that avoids Java will likely look elsewhere, since the codebase is Java and the documented paths are Docker or Ansible deployment. The README is a short feature-oriented document that points to external documentation for installation and usage detail, and it states no numbered release, so anyone evaluating it should read the readthedocs site rather than rely on the repository landing page alone.