Getting Started with SeekDB: An AI‑Native Embedded Search Database

Overview

SeekDB is an AI‑native search database that unifies vector, text, structured, and semi‑structured data in a single engine, enabling hybrid retrieval. Official repository: https://github.com/oceanbase/seekdb

Key Features

Embeddable like SQLite – no separate server process.

Full‑text search comparable to Elasticsearch.

Vector (embedding) search with built‑in embedding function.

Supports structured and semi‑structured queries.

Installation (requires uv )

uv init seekdb-demo
cd seekdb-demo
uv python pin 3.12
uv venv
uv pip install pyseekdb

Run the demo:

uv run src/demo.py

Complete Python Example (src/demo.py)

import pyseekdb
from pyseekdb import DefaultEmbeddingFunction, HNSWConfiguration

client = pyseekdb.Client()
coll_name = "notes"
collection = client.create_collection(name=coll_name)
print("Collection created:", collection.name)

docs = [
    "今天跑了 10 公里，状态不错。",
    "Mapbox 的 polyline 编码可以有效减少地址长度。",
    "Python asyncio 和 gevent 的对比需要实测。",
]
ids = ["run1", "map1", "py1"]
metas = [{"tag": "run"}, {"tag": "map"}, {"tag": "python"}]

collection.add(ids=ids, documents=docs, metadatas=metas)
print("Documents added.")

query = "跑步 状态"
res = collection.query(query_texts=query, n_results=2)
print("Query:", query)
for i, rid in enumerate(res["ids"][0]):
    print("Result", i + 1)
    print("  id:", rid)
    print("  doc:", res["documents"][0][i])
    print("  meta:", res["metadatas"][0][i])
    print("  distance:", res["distances"][0][i])

client.delete_collection(coll_name)

Advantages

Embedded experience: works like a local SQLite file, ideal for edge devices or small teams.

Simple API flow: init → add → query → delete.

Automatic embedding: the SDK handles vectorization internally.

Hybrid search aligns with Retrieval‑Augmented Generation (RAG) and other AI workloads.

Known Limitations

Current resource footprint is modest but noticeable (approximately 1 CPU core and 1 GB RAM); optimization is pending.