SeekDB: The AI-Native Search Database for Agent Storage

SeekDB is an AI-native search database built by OceanBase — the team behind one of the world's most battle-tested distributed SQL engines. It unifies vector, full-text, and scalar search into a single MySQL-compatible engine, so your AI agent no longer needs three separate databases to remember, retrieve, and filter.

Why Agents Need a State Store

Agent workloads are continuous: write a memory, search it milliseconds later, fork an experiment, merge it back. Traditional databases choke on this pattern — vector DBs have no scalar filters, full-text engines have no vectors, and none of them support safe sandboxing. SeekDB attacks all three at once.

Performance That Matters

SeekDB's async Change Stream pipeline decouples writes from index building, and its two-level HNSW (incremental + snapshot) makes freshly written vectors immediately searchable. The result: 1,523 QPS streaming write+search with 21.7 ms P99 — 10.7× the QPS of Milvus and 3.2× Elasticsearch, with P99 jitter of just 1.1× under concurrency.

Copy-on-Write Sandboxes

FORK DATABASE snapshots an entire database in seconds with no data copy. Agents experiment freely — write, query, even break tables — then MERGE TABLE commits the work back, or DROP DATABASE discards it. This is kernel-level COW, not application-layer save/restore.

Ecosystem Ready

MySQL protocol means full ACID and instant compatibility with LangChain, LlamaIndex, and Dify. Run it embedded (in-process), as a server, or on OceanBase for cloud scale. Apache-2.0 licensed with an active 2.9K-star community.