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

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).

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:

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:

from veadk import Agent
import asyncio

agent = Agent()

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

AgentKit application

Use the shared AgentKit application factory when your project needs AgentKit APIs, VeADK's bundled Web UI, health checks, and agent-topology endpoints. This keeps platform routes and lifecycle code out of your agent module:

from veadk import Agent
from veadk.integrations.agentkit import create_agentkit_app

root_agent = Agent(name="customer_support")
app = create_agentkit_app(root_agent)

See examples/generated_agentkit_project for a complete generated project.

The Agent Server metadata endpoint reports the root Agent's name, description, model, sub-Agents, tools, skills, and mounted component summaries. Each Runtime row in Studio has explicit connect and info actions; the info panel's tabs switch between this live metadata and control-plane information without exposing prompts or credentials. The same metadata advertises mounted smart-search sources, so Studio can disable unavailable sources up front and query the Agent's web-search tool, KnowledgeBase, or long-term memory without exposing component credentials. Studio also provides an isolated Insight Sandbox for temporary Codex conversations. It reuses a dedicated AgentKit CodeEnv tool, creates a fresh user-owned Sandbox session, and deletes that session on exit without adding the conversation to normal Studio history. Reloading may create another temporary session; AgentKit reclaims abandoned sessions automatically when their TTL ends. When configuring skills, Studio can also browse account-scoped AgentKit Skill Spaces and their paginated skill lists by region and project. These requests are signed on the server, so browser clients never receive Volcengine credentials.

The Studio deployment flow lists Feishu, knowledge-base, short-/long-term memory, and observability settings in their feature sections. Values entered there are mirrored in the deployment environment-variable summary and converted to VeADK runtime environment variables only when deploying; secrets are not written to generated source or exported YAML. For multi-instance runtimes, use a database-backed short-term memory store so sessions remain available across instances.

When a cloud image build fails from the bundled Web UI, the deployment error includes a credential-safe excerpt from the build log so dependency and Dockerfile failures can be diagnosed directly.

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.

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]):

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. Its custom-agent workbench supports in-page debugging followed by source review and AgentKit deployment configuration. Knowledge-base, memory, tool, and tracing components request only settings that cannot be derived automatically. Studio forwards its server-side Volcengine credentials and lets VeADK resolve Ark, embedding, media, speech, VeSearch, and APMPlus keys for debug runs and deployed runtimes:

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):

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:

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 studio deploy automatically provisions ServerlessApplicationRole when the required VeFaaS role is missing.

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 and forward its validated OAuth access token when connecting to an AgentKit runtime protected by custom_jwt; the login footer uses AgentKit product branding, and you can customize the browser/sidebar branding with --site-title and --site-logo (omitting the title keeps VeADK Studio)
• veadk studio deploy: deploy Studio and ensure its default IAM role has the required model, observability, search, security, memory, and identity system policies; target cn-beijing (default) or cn-shanghai with --region, automatically locate the Identity user pool across Beijing and Shanghai, and select the VeFaaS project with --project (default default); custom local or remote logo images are bundled into the deployment; the deployed client skips the second OAuth consent confirmation after login; two dedicated AgentKit CodeEnv Tools are created automatically for temporary chats and Skill creation unless their IDs are supplied with --sandbox-chat-codex-tool-id and --sandbox-skill-creator-tool-id
• veadk studio update --vefaas-app-name <app-name>: build the frontend from a local VeADK source checkout and release it through the existing VeFaaS Application and Function. Omit --region and --project to search Beijing, Shanghai, and all visible projects. Existing URL, SSO, IAM, gateway, environment variables, title, and logo are preserved; pass --site-title or --site-logo only when those branding values should be replaced. Sandbox Tool IDs are also preserved unless the corresponding deploy option is supplied

Studio can assign comma-separated local usernames or OAuth emails to the admin and developer roles:

veadk studio \
  --admin "admin,admin@example.com" \
  --developer "alice,alice@example.com"

veadk studio deploy \
  --user-pool-id <pool-id> \
  --allowed-client-id <client-id> \
  --vefaas-app-name <app-name> \
  --admin "admin@example.com" \
  --developer "alice@example.com,bob@example.com"

Omitting both role lists makes every signed-in user an admin. Supplying either list enables role-based access control; users not in a list are regular users, and admin wins when an identity appears in both lists. Admins have all Studio capabilities and can see every Runtime. Developers can add and manage agents but can see and manage only their own Runtimes. Regular users can see only their own Runtimes; the add/manage-agent sidebar items are hidden. The sidebar account footer shows the OAuth email beneath the display name and identifies the current role with a color-coded badge. The manage-agent view defaults to Beijing and can be switched to Shanghai. A new conversation renders the first message immediately while its server session initializes in the background. Existing Runtimes without owner metadata are visible only to admins unless they carry the legacy-compatible veadk:author ownership tag. Local usernames are browser-provided and can be impersonated, so use OAuth or gateway authentication for production authorization.

Contribution

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

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:

pytest -n 16

Security and privacy

This project takes security seriously. For vulnerability reporting and supported versions, see SECURITY.md

Contact with us

Join our discussion group by scanning the QR code below:

License

This project is licensed under the Apache 2.0 License.