OceanBase Releases seekdb: An Open Source AI Native Hybrid Search Database for Multi-model RAG...

OceanBase Releases seekdb: An Open Source AI Native Hybrid Search Database

OceanBase has released seekdb, an open-source AI-focused database under the Apache 2.0 license. seekdb is an AI-native search database that unifies relational data, vector data, text, JSON, and GIS in one engine and exposes hybrid search and in-database AI workflows.

What is seekdb?

seekdb is a lightweight, embedded version of the OceanBase engine, aimed at AI applications rather than general-purpose distributed deployments. It runs as a single-node database, supports embedded mode and client or server mode, and remains compatible with MySQL drivers and SQL syntax.

Key Features of seekdb

• Embedded database supported
• Standalone database supported
• Distributed database not supported
• Relational data with standard SQL
• Vector search
• Full text search
• JSON data
• Spatial GIS data
• Hybrid search as the core feature

Hybrid Search

The main feature OceanBase pushes is hybrid search. This is search that combines vector-based semantic retrieval, full-text keyword retrieval, and scalar filters in a single query and a single ranking step. seekdb implements hybrid search through a system package named DBMS_HYBRID_SEARCH with two entry points:

• DBMS_HYBRID_SEARCH.SEARCH which returns results as JSON, sorted by relevance
• DBMS_HYBRID_SEARCH.GET_SQL which returns the concrete SQL string used for execution

Vector and Full Text Engine Details

At its core, seekdb exposes a modern vector and full-text stack. For vectors, seekdb:

• Supports dense vectors and sparse vectors
• Supports Manhattan, Euclidean, inner product, and cosine distance metrics
• Provides in-memory index types such as HNSW, HNSW SQ, HNSW BQ
• Provides disk-based index types including IVF and IVF PQ

For text, seekdb offers full-text search with:

• Keyword, phrase, and Boolean queries
• BM25 ranking for relevance
• Multiple tokenizer modes

AI Functions Inside the Database

seekdb includes built-in AI function expressions that let you call models directly from SQL, without a separate application service mediating every call. The main functions are:

• AI_EMBED to convert text into embeddings
• AI_COMPLETE for text generation using a chat or completion model
• AI_RERANK to rerank a list of candidates
• AI_PROMPT to assemble prompt templates and dynamic values into a JSON object for AI_COMPLETE

Multimodal Data and Workloads

seekdb is built to handle multimodal data and workloads, including:

• Relational data
• Vector data
• Text data
• JSON data
• Spatial GIS data

This makes seekdb an ideal choice for applications that require a unified storage and indexing layer for multiple data types.