Quickstart

This tutorial explains how to configure a model and run a minimal agent that returns an answer in your terminal.

Complete the installation and obtain a Volcengine Ark API key first. This chapter covers only the minimal path to a first run; deploying to AgentKit, memory, knowledge bases, multi-agent, and evaluation are covered in their respective chapters.

Configure a model

Provide the credentials for calling the model in one of two ways. The simplest is an environment variable — only the API key is required.

Environment variable
Config file

With the API key set, VeADK uses the default doubao model on Volcengine Ark:

export MODEL_AGENT_API_KEY="your-ark-api-key"

To use a different model, also set MODEL_AGENT_NAME, MODEL_AGENT_PROVIDER, and MODEL_AGENT_API_BASE.

Create config.yaml in the project root; VeADK reads it automatically:

config.yaml

model:
  agent:
    provider: openai
    name: doubao-seed-1-8-251228
    api_base: https://ark.cn-beijing.volces.com/api/v3/
    api_key: # your Volcengine Ark API key

provider: openai uses the OpenAI-compatible protocol, and api_base points at Ark’s compatible endpoint, so Ark models such as doubao can be used directly.

Write your first agent

Create main.py:

main.py

from veadk import Agent
import asyncio

agent = Agent()

res = asyncio.run(agent.run("Introduce Volcengine in one sentence."))
print(res)

Agent holds the model and instruction; with no arguments it uses the default model from config.yaml. agent.run(...) is asynchronous and returns the final text answer.

Run

python main.py

The terminal prints the model’s answer, completing a minimal VeADK agent run.

Overview

Get started

Core components

Observability

Extensions

Frontend

Security

Configure a model

Write your first agent

Run

​Configure a model

​Write your first agent

​Run

Configure a model

Write your first agent

Run