# seekdb

> Open-source AI-native search database that unifies relational, vector, full-text, JSON and GIS data in one engine with hybrid search and in-database AI functions.

seekdb is an open-source, AI-native search database from the OceanBase team. It combines relational, vector, full-text, JSON and GIS data in a single engine, and is aimed at agent storage, RAG, and semantic search workloads. It can run embedded in a Python process or as a client/server database, and the repository is licensed under Apache 2.0.

## What It Is

seekdb is a MySQL-compatible database built on the OceanBase SQL engine, with full ACID support. It lets developers combine vector search, full-text search and relational filters in one SQL statement instead of merging results in client code. The Python SDK, pyseekdb, installs with pip and exposes a collection-style API for inserting documents and querying them.

## Search and In-Database AI

Hybrid search runs multi-path retrieval in a single query, with reranking by weighted scores, Reciprocal Rank Fusion, or LLM-based reranking. Vector support covers dense and sparse vectors, several distance metrics, and in-memory (HNSW variants) and disk-based (IVF, IVF_PQ) indexes. Full-text search offers keyword, phrase and Boolean queries with BM25 ranking. SQL functions such as AI_EMBED, AI_COMPLETE and AI_RERANK, managed through the DBMS_AI_SERVICE package, run embedding, text generation and reranking inside the database and can register external LLM services.

## Agent State Store

The README positions seekdb as a state store for AI agents. An asynchronous index pipeline decouples writes from index builds so newly written vectors can be searched right away. The README reports 1,523 QPS in its own streaming write-plus-search benchmark and compares it with Milvus and Elasticsearch. Copy-on-write sandboxes use FORK DATABASE to snapshot a database, then MERGE TABLE to accept changes or DROP DATABASE to discard them.

## Deployment and Ecosystem

Options include the embedded Python library, Docker, a Linux install script, Homebrew on macOS, building from source, and a cloud instance created via an API call. The README lists integrations with LangChain, LlamaIndex, Dify, LangGraph, Coze and HuggingFace. The homepage also promotes seekdb D0 (a database for agents, MCP and RAG) and seekdb M0 (cloud memory for OpenClaw).

## Features
- Hybrid search combining vector, full-text and relational filters in one SQL statement
- Dense and sparse vectors with HNSW, HNSW_SQ, HNSW_BQ, IVF and IVF_PQ indexes
- Full-text search with BM25 ranking
- Reranking via weighted scores, RRF or LLM
- In-database AI functions: AI_EMBED, AI_COMPLETE, AI_RERANK
- Copy-on-write FORK/MERGE database sandboxes
- Embedded and client/server deployment modes
- MySQL protocol compatibility with ACID transactions
- Multi-model storage: relational, vector, text, JSON, GIS
- Python SDK (pyseekdb)

## Integrations
LangChain, LlamaIndex, Dify, LangGraph, Coze, HuggingFace, Camel-AI, DB-GPT, FastGPT, Firecrawl, Spring-AI-Alibaba, Cloudflare Workers AI, Jina AI, Ragas, Instructor, Baseten, OceanBase

## Platforms
LINUX, MACOS, API, DEVELOPER_SDK, CLI

## Pricing
Open Source

## Version
v1.4.0

## Links
- Website: https://seekdb.ai
- Documentation: https://docs.seekdb.ai/docs/
- Repository: https://github.com/oceanbase/seekdb
- EveryDev.ai: https://www.everydev.ai/tools/seekdb