Volcengine Agent Development Kit

Project Positioning#

VeADK is an official open-source Agent Development Kit from Volcengine, primarily developed in Python (95%). It serves as a "glue layer" and development standard layer connecting Volcengine infrastructure. The project aims to address enterprise AI application development challenges including complex model integration, fragmented memory/knowledge management, lack of tool calling standards, and lengthy deployment-to-production pipelines.

Core Features:

• Multi-Ecosystem Compatibility: Full compatibility with Google ADK for seamless project migration; LiteLLM integration for mainstream model access; OpenAI API compatible model services
• Memory & Knowledge: Short-term memory via MySQL/PostgreSQL persistence; Long-term memory through Viking DB and cloud search services with vector retrieval; Knowledge base with LlamaIndex as core processing entry
• Rich Toolchain: Web Search, image/video generation, code sandbox, Feishu Lark integration, LAS AI data lake
• Protocol Support: Native MCP (Model Context Protocol) and A2A (Agent-to-Agent) protocol support
• Cloud-Native Deployment: Integration with VeFaaS function compute, API Gateway, CloudEngine; Docker image and code package one-click deployment
• Enterprise Security: AgentKit Identity integration, IdP (SAML/OIDC) support, third-party credential management, attribute-based dynamic authorization
• Observability: CozeLoop, APMPlus, TLS integration with Tracing and online evaluation

Architecture#

Foundation: Built on google-adk (>=1.19.0), reusing Google ADK's agent architecture design

Inference Layer: Unified model inference encapsulation via litellm (>=1.74.3), connecting to Volcengine ARK or other OpenAI-compatible services

Protocol Layer: a2a-sdk (Agent2Agent) and mcp (Model Context Protocol) for cross-agent and tool interactions

Memory Management: Layered design with psycopg2/pymysql for relational data (short-term) and vikingdb-python-sdk for vector data (long-term)

Installation & Usage#

Requirements: Python >= 3.10 (3.12 recommended)

Installation:

# PyPI stable
pip install veadk-python

# With extensions
pip install veadk-python[extensions]

# From source
uv venv --python 3.12
uv sync
uv pip install -e .

# Docker
veadk-cn-beijing.cr.volces.com/veadk/veadk-python:latest

Quick Example:

from veadk import Agent
import asyncio

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

CLI Tools:

• veadk init: Initialize demo project
• veadk deploy: Deploy to Volcengine VeFaaS
• veadk prompt: Optimize prompts via PromptPilot

Associated Services#

• Compute: VeFaaS Function Compute
• Models: ARK Model Inference Platform (doubao series)
• Data: LAS AI Data Lake, Viking DB Vector Database
• Tools: PromptPilot prompt optimization, CozeLoop observability

License#

Apache-2.0 License

Current Version#

0.5.22 (actively developed as of research time)