Configure AK/SK credentials first, or obtain short-lived STS credentials through SSO login.
launch, deploy, and invoke use the same credential precedence to resolve valid credentials. See the Authentication section of the Quickstart. For the full guide to writing agent code, see the VeADK docs.1
Scaffold a project
Create a project from a built-in template. Run
agentkit init --list-templates to list every template first.2
Edit the agent
Modify the scaffolded code to adjust the agent’s instruction, tools, and orchestration logic.
The runtime container must listen on
0.0.0.0:8000 to pass its readiness probe. Templates satisfy this by default, so you rarely need to change it.3
Configure the model and cloud
Follow the Quickstart to select Volcengine or BytePlus and configure the model name, API base URL, and API key. Deployment credentials and model credentials serve different purposes; CLI login alone does not configure the modelAdd This writes actual values to local
agentkit.yaml and .env to .gitignore and .dockerignore in the project root, then write the exported model variables to lifecycle configurationagentkit.yaml; root configuration does not substitute ${VAR} inside the file4
Build and deploy
Run the full lifecycle flow, building the image in the cloud and creating or updating the runtime.
5
Invoke the runtime
Read the current project’s
agentkit.yaml and invoke the runtime to verify the deployment.6
Inspect and iterate
Read the logs when the runtime misbehaves; edit the code and redeploy to publish a new version.To roll back or adjust resources, use the version and update commands: list history with
agentkit runtime versions my-agent, publish a specific version with agentkit runtime release my-agent --rev <n>, and change CPU and memory with agentkit runtime update my-agent --cpu-milli 1000 --memory-mb 2048 --auto-release.