Volcengine Agent Development Kit

An open-source kit for agent development, integrated the powerful capabilities of Volcengine.

For more details, see our documents.

A tutorial is available by Jupyter Notebook, or open it in Google Colab directly.

Installation

From PyPI

python
pip install veadk-python

# install extensions
pip install veadk-python[extensions]

Build from source

We use uv to build this project (how-to-install-uv).

bash
git clone ... # clone repo first

cd veadk-python

# create a virtual environment with python 3.12
uv venv --python 3.12

# 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 .

Configuration

We recommand you to create a config.yaml file in the root directory of your own project, VeADK is able to read it automatically. For running a minimal agent, you just need to set the following configs in your config.yaml file:

yaml
model:
  agent:
    provider: openai
    name: doubao-seed-1-6-250615
    api_base: https://ark.cn-beijing.volces.com/api/v3/
    api_key: # <-- set your Volcengine ARK api key here

You can refer to the config instructions for more details.

Have a try

Enjoy a minimal agent from VeADK:

python
from veadk import Agent
import asyncio

agent = Agent()

res = asyncio.run(agent.run("hello!"))
print(res)

Feishu bot channel

VeADK now provides veadk.extensions.FeishuChannelExtension for bridging a Feishu bot with a Runner. It maps union_id to user_id, and thread_id / chat_id to session_id, so VeADK memory and tracing can work directly in Feishu conversations.

python
from veadk import Agent, Runner
from veadk.extensions import FeishuChannelExtension

agent = Agent()
runner = Runner(agent=agent, app_name="feishu_demo")
channel = FeishuChannelExtension(runner=runner)

Configure credentials with TOOL_FEISHU_CHANNEL_APP_ID and TOOL_FEISHU_CHANNEL_APP_SECRET, or in config.yaml under tool.feishu_channel.

A2UI (agent-driven UI)

VeADK integrates Google's A2UI, letting an agent reply with
declarative UI (cards, rows, forms) instead of plain text. A client renders the
UI with native components. Enable it with a single flag (requires the optional
a2ui-agent-sdk dependency: pip install veadk-python[a2ui]):

python
from veadk import Agent

agent = Agent(enable_a2ui=True)  # uses the bundled "basic" component catalog

A bundled React web UI renders A2UI over the standard ADK API server. The built
UI ships inside the package (veadk/webui, produced by npm run build), so
installed users can launch it directly:

bash
veadk frontend --agents-dir examples           # serve UI + API on http://127.0.0.1:8000

To rebuild the UI from source (output goes to veadk/webui, which is committed
so it ships with the wheel):

bash
cd frontend && npm install && npm run build

Point the agent at a custom component catalog (relative paths resolve against the
agent's directory; absolute paths work too). With no argument it auto-discovers a
catalog.json next to the agent, falling back to the bundled basic catalog:

python
Agent(enable_a2ui=True, a2ui_catalog="catalog.json")  # beside the agent

Enterprises extend the component set in two matching halves: a backend catalog
(a catalog.json or a veadk.a2ui.BaseA2UICatalog subclass) and a frontend
renderer directory (frontend/src/a2ui/components/<Name>/). See
frontend/README.md.

Command line tools

VeADK provides several useful command line tools for faster deployment and optimization, such as:

veadk deploy: deploy your project to Volcengine VeFaaS platform (you can use veadk init to init a demo project first)
veadk prompt: otpimize the system prompt of your agent by PromptPilot
veadk frontend: serve the A2UI web UI together with the ADK agent API server

Contribution

Before making your contribution to our repository, please install and config the pre-commit linter first.

bash
pip install pre-commit
pre-commit install

Before commit or push your changes, please make sure the unittests are passed ,otherwise your PR will be rejected by CI/CD workflow. Running the unittests by: