AgentKit Samples

Official sample repository for ByteDance's Volcengine AgentKit platform, demonstrating best practices for multi-agent collaboration, RAG memory management, and cloud-native deployment using the VeADK framework.

Overview#

AgentKit Samples is the official sample repository for ByteDance's Volcengine AgentKit platform. Built on VeADK (Volcengine Agent Development Kit), it provides comprehensive solutions for enterprise-grade AI agent development. Maintained by ByteDance/Volcengine team under Apache-2.0 license.

Core Capabilities#

Multi-Agent Collaboration#

Hierarchical structure with specialized division of labor
A2A (Agent-to-Agent) protocol for distributed collaboration
Examples: multi_agents, a2a_simple

Memory System#

Short-term Memory: Session-level context persistence via MySQL/PostgreSQL
Long-term Memory: Vector retrieval via VikingDB and cloud search services
Example: vikingmem_agent, Tutorial 04-Memory

Knowledge Base Integration (RAG)#

LlamaIndex data processing pipeline
Viking knowledge base backend
Cloud search service integration
Example: vikingdb_agent

Tool Ecosystem#

MCP (Model Context Protocol) integration
Built-in Web Search, code sandbox, image/video generation
Examples: mcp_simple, various use-cases

Observability#

Call chain Tracing
Log service (TLS) storage and retrieval
APMPlus, CozeLoop integration
Online evaluation capabilities
Tutorial: 06-Observability

Cloud-Native Deployment#

Docker containerization
VeFaaS one-click deployment
API Gateway integration
Command: veadk deploy

Identity & Permissions#

Inbound authentication
OAuth2/API Key outbound authentication
Permission policy management
Tutorial: 03-AgentKit-Identity

Typical Use Cases#

Scenario	Description	Example Path
Basic Chat Agent	Simple dialogue with short-term memory	hello_world
Travel Planner	Web search + domain knowledge for trip planning	travel_planner
Restaurant Ordering	Complex workflows, async tool calls	restaurant_ordering
Real-time Voice Chat	Python server + Web client	realtime_voice
AI Coding Assistant	Code writing and optimization	ai_coding
Customer Support Agent	Post-sales consultation & pre-sales guidance	customer_support
Video Generation	Multi-tool video content creation	video_gen
Store Inspection	Multi-agent collaborative inspection	store_inspection
Data Analysis	LanceDB-based analytics	data_analysis_with_datalake
E-commerce Marketing Video	A2A multi-agent video creation	ecommerce_video

Quick Start#

Requirements: Python 3.10+

Install SDK:

pip install veadk-python
pip install veadk-python[extensions]  # Optional extensions

Configuration (config.yaml):

model:
  agent:
    provider: openai
    name: doubao-seed-1-6-250615
    api_base: https://ark.cn-beijing.volces.com/api/v3/
    api_key: # Your Volcengine ARK API Key

Minimal Example:

from veadk import Agent
import asyncio

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

Project Structure#

├── 01-tutorials/        # Step-by-step tutorials
├── 02-use-cases/        # Advanced use cases
├── 03-integrations/     # Third-party integrations (lark_bot, etc.)
├── skills/              # Skill modules
├── template/            # Project templates
└── workflow_utils/      # Workflow utilities

CLI Commands#

veadk deploy: Deploy Agent to Volcengine VeFaaS platform
veadk prompt: Optimize system prompts via PromptPilot

Dependency Ecosystem#

Core SDK: veadk-python, agentkit-sdk-python
Cloud Services: Volcengine ARK (model inference), VeFaaS (function compute), VikingDB (vector DB), TLS (logging), CozeLoop (observability)
External Tools: LlamaIndex, MCP, Docker

Language Distribution#

Python (72.5%), Jupyter Notebook, HTML, JavaScript, Go