Changelog
2026.1.30: New releases of seekdb V1.1.0, pyseekdb V1.1.0, and seekdb-js V1.1.0
seekdb V1.1.0
- macOS support: seekdb can now be built natively and used for local development on macOS 15 and later. For installation on Mac, see Deploy seekdb with package managers.
- FORK TABLE: New
FORK TABLEfeature provides fast table copies without full data duplication. It reuses storage at the table level and uses copy-on-write, so copies are quick and light while keeping consistency. This replaces the need for full CTAS-style copies in many cases. - Storage: Lower minimum resource requirements for data storage, clog storage, and system log storage, making it easier to run seekdb in resource-constrained environments.
- Memory management: Memory handling has been improved for different deployment sizes (large or small memory, standalone or co-located) so that behavior is more predictable across environments.
For full details, see the seekdb V1.1.0 Release Note.
pyseekdb V1.1.0
V1.1.0 adds the following:
- More flexible collection names: Maximum name length is now 512 characters (up from 64), so you can use longer, more descriptive names.
- Full-text search: New
FulltextIndexConfiglets you configure custom tokenizers and field mappings. - Broader embedding support: Integration with 11 popular embedding platforms, including open-source and commercial models.
- Persistent metadata: Embedding function definitions can be stored in system tables for easier setup and reuse.
- table-fork (experimental): Snapshot-style copy at the collection level for experiments and rollback.
For full details, see the pyseekdb changelog.
seekdb-js V1.1.0
V1.1.0 adds the following:
- More flexible collection names: Maximum name length is now 512 characters (up from 64), so you can use longer, more descriptive names.
- Full-text search: New
FulltextAnalyzerConfiglets you configure custom tokenizers and field mappings. - Broader embedding support: Integration with 11 popular embedding platforms, including open-source and commercial models.
- Persistent metadata: Embedding function definitions can be stored in system tables for easier setup and reuse.
- table-fork (experimental): Snapshot-style copy at the collection level for experiments and rollback.
For full details, see the seekdb-js changelog.
2025.12.29: New release of seekdb V1.0.1
This release focuses on vector index improvements: the hnsw_bq index type now supports IP (inner product) and cosine distance metrics, and the upper bound for vector search has been raised. It also includes fixes for vector indexing, semantic indexing, and vector search.
For full details, see the seekdb V1.0.1 Release Note.
2025.11.14: New releases of seekdb V1.0.0 and pyseekdb V1.0.0
seekdb V1.0.0
OceanBase seekdb (seekdb) is an AI-native hybrid search database. It unifies vectors, full text, and structured and semi-structured data in one engine and supports multi-modal hybrid search and inference through built-in AI functions.
For full details, see the seekdb V1.0.0 Release Note.
Highlights:
- Quick to build: Get an AI application running with a few lines of Python. The stack is light enough to run VectorDBBench on 1 vCPU and 2 GB RAM.
- Hybrid search: Combine vector search, full-text search, and scalar filters in a single query, with support for multi-stage pipelines (e.g. coarse ranking then fine ranking).
- Multi-model data and indexing: Supports scalars, vectors, full-text, JSON, and GIS data and their corresponding indexes within a single engine.
- AI inside the database: Embeddings, inference, prompt management, and reranking run in the database, supporting a full document-in/data-out RAG workflow.
- SQL-native: Built on OceanBase, with real-time writes and reads, ACID transactions, and MySQL compatibility.
pyseekdb V1.0.0
pyseekdb is the official Python client for OceanBase. It uses a single API with two connection modes: one for embedded seekdb (in-process) and one for a remote server (server-mode seekdb or OceanBase).
For an introduction and examples, see pyseekdb quick start.