Project Structure¶

agentengine init creates a small project that can be run locally and later packaged for deployment.

Typical Layout¶

my-agent/
  agent.py
  agentengine.yaml
  .env

Some projects use a package layout:

my-agent/
  my_agent/
    __init__.py
    agent.py
  agentengine.yaml
  .env

KsADK also supports ksadk.yaml and ksadk.yml for compatibility.

`agent.py`¶

The agent entry module should export the object configured as agent_variable. The default variable is root_agent.

root_agent = graph.compile()

For ADK projects this is usually a google.adk.agents.Agent. For LangGraph it is commonly a compiled graph. For LangChain it may be a runnable chain. For DeepAgents it is the object returned by create_deep_agent.

`agentengine.yaml`¶

The project YAML makes framework detection explicit.

name: my-agent
framework: langgraph
entry_point: agent.py
agent_variable: root_agent

Supported public framework values include:

adk
langchain
langgraph
deepagents

`.env`¶

The .env file is for local secrets and provider settings. Keep it out of Git.

OPENAI_API_KEY=sk-test
OPENAI_BASE_URL=https://api.example.com/v1
OPENAI_MODEL_NAME=my-model

For public docs and examples, use placeholders only. Do not publish real tokens, internal endpoints, private registry names, cookies, kubeconfig paths, or customer data.

Generated Files¶

Local runs may create caches, virtual environments, build output, or runtime state. These are not source files:

.venv/
__pycache__/
.pytest_cache/
dist/
build/
site/
.agentengine/
.agentengine.state

Importing An Existing Agent¶

Use --from-agent when you already have a Python file or directory:

agentengine init my-agent --from-agent ./existing_agent.py
agentengine init my-agent --from-agent ./existing_agent_dir

After import, inspect agentengine.yaml and confirm the detected framework, entry point, and exported variable before running the project.