Seekdb，unified search database for AI(relational, vector and full text) (github.com)

OceanBase announced SeekDB, an AI-native, open‑source (Apache 2.0) search database that unifies relational, vector, full‑text, JSON and GIS data in a single engine to enable hybrid search and in‑database AI workflows. SeekDB is MySQL‑compatible and ACID‑compliant, supports embedded and single‑node deployments (with an OceanBase server mode for distributed use), and aims to cut integration overhead by letting teams run embeddings, reranking, LLM inference and prompt management inside the database for end‑to‑end RAG/document‑in→data‑out pipelines. Technically, SeekDB exposes VECTOR columns (default embedding dim 384) and vector indexes, lets you combine vector similarity and full‑text matches in one SQL statement, and provides a Python SDK (pyseekdb) that auto‑generates embeddings via a DefaultEmbeddingFunction. It ships as a binary, Docker image or embedded library, and is touted as lightweight enough to run VectorDBBench on a 1C/2G machine. Key implications: developers get lower latency and simpler architectures (no separate VectorDB + SQL store), smoother migration for existing MySQL ecosystems, and a unified substrate for semantic search, agent memory, edge deployments and AI‑assisted tooling.