Docker image based on python docker image with Google Agent Development Kit (ADK) pre-installed for streamlined AI agent development.
This image is a base image for building your own AI agents using the ADK, all it is doing is providing a base image with the ADK pre-installed. It is not meant to be run directly, but rather to be extended in your own agent project for rapid development.
- Base:
python:3.12 - Workdir:
/app - Deps: copies in
requirements.txt(pins version ofgoogle-adkmatching the version of this image)
You won’t run adk-docker-base by itself. Instead, extend it in your own agent project with your ADK Python code. If you need to install any extra dependencies, you can do so in your own Dockerfile which means using this immage is probably not a good idea. The intent of this image here is to allow for rapid iteration of ADK based agents off of a common Docker base image.
Follow the ADK quickstart structure:
├── Dockerfile
└── my_agent/
├── __init__.py
├── agent.py
└── .env
from . import agentfrom google.adk.agents import Agent
def echo_tool(text_to_echo: str) -> dict:
"""Echoes the provided text back.
Args:
text_to_echo (str): The text to echo.
Returns:
dict: status and the echoed content.
"""
return {"status": "success", "report": f"Echo: {text_to_echo}"}
root_agent = Agent(
name="echo_agent",
model="gemini-2.5-flash", # Pick the best model for your use case
description="An agent that echoes back the user's input.",
instruction="Echo back whatever the user says.",
tools=[
echo_tool
],
)If you are actually going to use this image somewhere, don't actually stick with latest, specify an actual version so it is consistent.
FROM ghcr.io/unitvectory-labs/adk-docker-base:latest
WORKDIR /app
COPY my_agent/ ./my_agent/
EXPOSE 8000
# Run the Web UI for ADK
CMD ["adk", "web", "--host", "0.0.0.0", "--port", "8000"]
podman build -t adk-example-agent .
podman run --rm -p 8000:8000 --name adk-example-agent adk-example-agent