Volcengine Agent Development Kit (VeADK)

What it is

• A Python library for building AI agents that run on Volcengine’s large‑model platform (ARK). It bundles a lightweight runtime, tool‑integration helpers, and optional web UI/Studio components.

Key capabilities

• Agent abstraction – veadk.Agent wraps a language model (e.g., OpenAI‑compatible models hosted on Volcengine) and provides a simple run(prompt) async API.
• Tool extensions – Built‑in support for Feishu bot channels, database‑backed short‑term memory, evaluation tools, and a CLI.
• AgentKit integration – One‑line factory (create_agentkit_app) produces a FastAPI app exposing health checks, metadata, and Studio‑compatible endpoints (dynamic tools, web‑search, knowledge‑base, long‑term memory, etc.).
• Configuration‑by‑YAML – A config.yaml at the project root can specify model provider, API endpoint, credentials, and optional tool settings.
• Developer resources – Jupyter tutorial, Google‑Colab link, and comprehensive docs.

Getting started

# Install from PyPI (core + optional extensions)
pip install veadk-python[extensions]

from veadk import Agent
import asyncio

agent = Agent()  # reads config.yaml automatically
result = asyncio.run(agent.run("hello!"))
print(result)

Typical workflow

1. Create a config.yaml with your Volcengine ARK API key and model name.
2. Instantiate Agent (or a named sub‑agent) in your code.
3. Optionally wrap it with create_agentkit_app to expose a FastAPI server that Studio can manage (metadata, tool routing, feedback collection).
4. Deploy the server or run locally; use the provided Feishu channel extension to connect the agent to a Feishu bot.

Extensibility

• Database extra – uv sync --extra database adds persistent short‑term memory.
• Eval extra – adds evaluation set handling for feedback.
• CLI extra – provides command‑line utilities for running agents and managing sessions.

Community & support

• Open‑source under Apache 2.0.
• Documentation site: https://volcengine.github.io/veadk-python/
• Discussion group (QR code in repo) for real‑time help.
• Security policy and CI tests are enforced; contributions require passing pytest.

Who should use it

• Developers who want to prototype or ship LLM‑powered agents quickly on Volcengine’s infrastructure.
• Teams needing built‑in integrations (Feishu, knowledge bases, memory stores) without writing boilerplate networking code.
• Users of Volcengine’s AgentKit platform who want a Python SDK that aligns with the Studio UI.