安装
本章主要介绍 VeADK 的安装方法和基本配置项。
环境
PyPI
您可以直接使用pip从 PyPI 平台进行安装：
pip install veadk-python
从源码构建
本项目使用uv进行构建（安装uv）。
# clone repo first
git clone https://github.com/volcengine/veadk-python.git
cd veadk-python
uv venv --python 3.10
# Activate
# macOS/Linux: source .venv/bin/activate
# Windows CMD: .venv\Scripts\activate.bat
# Windows PowerShell: .venv\Scripts\Activate.ps1
# only install necessary requirements
uv sync
# or, install extra requirements
# uv sync --extra database
# uv sync --extra eval
# uv sync --extra cli
# or, directly install all requirements
# uv sync --all-extras
# install veadk-python with editable mode
uv pip install -e .
配置
VeADK 在仓库中提供了一个示例配置文件config.yaml.example，其中标明了一个智能体运行的必需（required）配置和可选（optional）配置。您可以基于该示例文件，在项目根目录下创建实际使用的配置文件config.yaml。VeADK 中的配置模块将自动查找并加载该文件内容，并将其中的配置项映射为运行时环境变量，从而帮助您省去环境配置的时间。
完整的配置项可以参考config.yaml.example文件。
说明
下面是详细的配置说明：
model:
# [required] for running agent
agent:
provider: openai # 如果使用的是方舟大模型，请在这里填入openai
name: doubao-1-5-pro-256k-250115
api_base: https://ark.cn-beijing.volces.com/api/v3/
api_key:
# [optional] for llm-as-a-judge a evaluation
judge:
name: doubao-1-5-pro-256k-250115
api_base: https://ark.cn-beijing.volces.com/api/v3/
api_key:
# [optional] for knowledgebase
embedding:
name: doubao-embedding-text-240715
dim: 2560
api_base: https://ark.cn-beijing.volces.com/api/v3/embeddings
api_key:
volcengine:
# [optional] for Viking DB and `web_search` tool in VolcEngine (https://console.volcengine.com/iam/keymanage/)
access_key:
secret_key:
tool:
# [optional] https://console.volcengine.com/ask-echo/my-agent
vesearch:
endpoint: # `bot_id`
api_key:
# [optional] https://www.volcengine.com/docs/84296/1494115
web_scraper:
endpoint:
api_key: # `token`
# [optional] https://open.larkoffice.com/app
lark:
endpoint: # `app_id`
api_key: # `app_secret`
token: # `user_token`
# [optional] for Volcengine Lake AI Service (https://www.volcengine.com/product/las)
las:
url: # mcp sse url
dataset_id: # dataset name
observability:
# [optional] for exporting tracing data to Volcengine CozeLoop and APMPlus platform
opentelemetry:
apmplus:
endpoint: http://apmplus-cn-beijing.volces.com:4317
api_key:
service_name:
cozeloop:
endpoint: https://api.coze.cn/v1/loop/opentelemetry/v1/traces
api_key:
service_name: # Coze loop `space_id`
tls:
endpoint: https://tls-cn-beijing.volces.com:4318/v1/traces
service_name: # TLS `topic_id`
region: cn-beijing
# [optional] for exporting evaluation data to Volcengine VMP (https://console.volcengine.com/prometheus)
prometheus:
pushgateway_url:
username:
password:
database:
# [optional]
opensearch:
host: # should without `http://` or `https://`
port: 9200 # default OpenSearch port
username:
password:
# [optional]
mysql:
host:
user:
password:
database:
charset: utf8
# [optional]
redis:
host:
port: 6379 # default Redis port
password:
db: 0 # default
# [optional] for knowledgebase (https://console.volcengine.com/vikingdb)
viking:
project: # user project in Volcengine Viking DB
region: cn-beijing
# [optional] for knowledgebase with viking database (https://console.volcengine.com/tos)
tos:
endpoint: tos-cn-beijing.volces.com # default Volcengine TOS endpoint
region: cn-beijing # default Volcengine TOS region
bucket:
# [optional] for prompt optimization in cli/app (https://www.volcengine.com/docs/82379/1587837)
agent_pilot:
api_key:
logging:
# ERROR
# WARNING
# INFO
# DEBUG
level: DEBUG
管理
为管理繁多的配置项，VeADK 提供了根据config.yaml文件的自动化配置管理方案。您在配置文件中的所有配置将会根据层级，自动转为大写并使用下划线连接，统一注册成为运行时的环境变量。
例如下面的配置项：
model:
name:
api_key:
api_base:
base_a:
base_b:
...
将会被转为如下几条环境变量：
MODEL_NAME=
MODEL_API_KEY=
MODEL_API_BASE_BASE_A=
MODEL_API_BASE_BASE_B=
...
VeADK 中提供了一个getenv方法来读取相关配置，您无需在各组件中次手动传入某个配置的参数。