Use OceanBase Vector Database in DB-GPT for knowledge base Q&A

Overview

DB-GPT is an open-source AI database assistant that supports using OceanBase as a vector database for knowledge base question-answering.

This topic describes how DB-GPT uses OceanBase vector search to store and retrieve document vectors, and integrates LLMs such as Qwen to deliver accurate question-answering services.

Version compatibility

• OceanBase Database: ≥ V4.3.3
• DB-GPT: v0.7.x and later

Prerequisites

Before integrating DB-GPT with OceanBase Database, ensure that:

• You have deployed OceanBase Database and created a MySQL user tenant.
• You have set the ob_vector_memory_limit_percentage parameter to enable vector search. We recommend setting this parameter to 30 in ODP versions earlier than V4.3.5 BP3, and keeping the default value of 0 in ODP V4.3.5 BP3 and later.
• Docker is deployed and running, and the current user has permission to execute Docker commands (verify with docker info).
• You have obtained an API key for Qwen. To obtain it, go to the Qwen API Key page.
• You have installed Git for the operating system you're using.

Step 1: Obtain the database connection string

You can use the following command to quickly deploy an independent OceanBase Database for testing:

docker run --name=ob433 -e MODE=slim -p 2881:2881 -e OB_TENANT_PASSWORD=****** -d quay.io/oceanbase/oceanbase-ce:latest

Wait for the container initialization to complete (about 2-3 minutes). You can check the status by running the following command. If the output is boot success!, the startup is successful.

docker logs ob433 | tail -1

The database connection string is as follows:

mysql -h 127.0.0.1 -uroot@test -P2881 -p********* -Dtest

Step 2: Install DB-GPT

This example uses source code deployment. For more deployment methods, see DB-GPT official website.

1. Download the source code

git clone https://github.com/eosphoros-ai/DB-GPT.git
cd DB-GPT

2. Install the uv package manager

# macOS / Linux
curl -LsSf https://astral.sh/uv/install.sh | sh
source $HOME/.local/bin/env
# Verify the installation by running the following command, which should output something like: uv 0.10.2
uv --version

3. Install the basic dependencies

For example, for Qwen, perform the following steps:

uv sync --all-packages \
  --extra "base" \
  --extra "proxy_tongyi" \
  --extra "rag" \
  --extra "storage_chromadb" \
  --extra "dbgpts"

4. Install the OceanBase partner package

uv pip install pyobvector

5. Connect to OceanBase and set memory usage ratio for vector data (optional)

source .venv/bin/activate
python3

from pyobvector import ObVecClient

client = ObVecClient(uri="127.0.0.1:2881", user="root@test",password="",db_name="test")
client.perform_raw_text_sql(
    "ALTER SYSTEM ob_vector_memory_limit_percentage = 30"
)

Step 3: Configure DB-GPT to use OceanBase Database

After the basic dependencies are downloaded, modify the TOML configuration file to switch to OceanBase Database.

Configure TOML file

In DB-GPT v0.7.x and later, we recommend using the TOML configuration file. Edit configs/dbgpt-proxy-tongyi.toml as follows:

[system]
# Load language from environment variable (It is set by the hook)
language = "${env:DBGPT_LANG:-en}"
api_keys = []
encrypt_key = "your_secret_key"

# Server Configurations
[service.web]
host = "0.0.0.0"
port = 5670

[service.web.database]
type = "sqlite"
path = "pilot/meta_data/dbgpt.db"

# Vector Store Configuration - Use OceanBase
[rag.storage]
[rag.storage.vector]
type = "OceanBase"
ob_host = "127.0.0.1"
ob_port = 2881
ob_user = "root@test"
ob_database = "test"
ob_password = ""
# ob_enable_normalize_vector = true

# Model Configurations
[models]

# Qwen LLM model
[[models.llms]]
name = "qwen-plus"
provider = "${env:LLM_MODEL_PROVIDER:-proxy/tongyi}"
api_base = "https://dashscope.aliyuncs.com/compatible-mode/v1"
api_key = "${env:DASHSCOPE_API_KEY}"

# Qwen Embedding model
[[models.embeddings]]
name = "text-embedding-v3"
provider = "${env:EMBEDDING_MODEL_PROVIDER:-proxy/tongyi}"
api_url = "https://dashscope.aliyuncs.com/compatible-mode/v1/embeddings"
api_key = "${env:DASHSCOPE_API_KEY}"

Parameter description:

• $ob_host: The IP address of the OceanBase Database.
• $ob_port: The port of the OceanBase Database.
• $ob_user: The username of the OceanBase Database.
• $ob_password: The password of the OceanBase Database user.
• $ob_database: The name of the OceanBase database.

Step 4: Start DB-GPT

First, set the DashScope API Key environment variable, then start DB-GPT:

# Set the Qwen API Key
export DASHSCOPE_API_KEY="your-dashscope-api-key"

# Start DB-GPT
uv run dbgpt start webserver --config configs/dbgpt-proxy-tongyi.toml

Once started successfully, visit http://localhost:5670 to access the DB-GPT web interface.

Step 5: Use the knowledge base for Q&A

1. Open your browser and go to http://localhost:5670.
2. Go to the Knowledge page and click Create Knowledge to create a knowledge base.
3. Fill in the Space Config and click Next.
4. Select the Datasource type and click Next.
5. Go to the Upload page, upload the file, and click Next.
6. Go to the Segmentation page and select the document slicing method.
7. The documents are sliced and vectorized and stored in the OceanBase vector database.
8. In the Explore section, select the knowledge space to perform Q&A.

Related documents/next steps

• DB-GPT official documentation
• DB-GPT KBQA FAQ