SeekDB Tutorial: Deploy Vector + Full-Text Search for AI Agents

In this tutorial, you'll deploy SeekDB — the AI-native search database from OceanBase — and run your first vector + full-text hybrid query. You can try it in 30 seconds with the Python SDK, or run it as a Docker server for production workloads.

Step 1: Try the Embedded Mode

No servers, no schemas, no embedding setup. Install the Python SDK and go:

pip install -U pyseekdb

Embedded mode runs in-process; switch to server or OceanBase mode with a single line of configuration.

Step 2: Fork a Sandbox for Agent Exploration

The signature feature: copy-on-write sandboxes. Snapshot your whole database in seconds — no data copy — and let your agent experiment freely:

FORK DATABASE agent_state TO agent_sandbox_42;
USE agent_sandbox_42;
INSERT INTO memory (session_id, embedding, content) VALUES (...);
-- Merge the good work back
MERGE TABLE agent_sandbox_42.memory TO agent_state.memory;

Step 3: Run One Hybrid Query

Vector, full-text, and scalar filters combine in a single SQL query — no separate databases, no manual reconciliation:

SELECT * FROM memory
WHERE MATCH(content) AGAINST ('budget report')
  AND category = 'finance'
ORDER BY VEC_DISTANCE(embedding, ?)
LIMIT 10;

Step 4: Deploy with Docker

For production, run the official image oceanbase/seekdb:latest (70K+ pulls). We recommend 2 CPU / 4 GB RAM minimum, and 4 CPU / 8 GB for real workloads. Use the pre-generated docker-compose on the tool page to get started in minutes.

Why It Works So Well

The async Change Stream pipeline decouples DML from index building, and two-level HNSW makes new vectors instantly searchable — 1,523 QPS with 21.7 ms P99, 10.7× Milvus and 3.2× Elasticsearch throughput. Full ACID and MySQL protocol mean LangChain, LlamaIndex, and Dify work without adapters.