🐢 Tiny Gordon

Tiny Gordon is an AI agent built in a flash using Docker Model Runner, Docker Compose, Google's ADK, and the MCP Toolkit to give it superpowers. This agent is designed to help you master Docker with ease.

🚧 this is a work in progress

Requirements

• Docker Desktop + MCP Toolkit extension
• Install the Brave MCP server (you need an API key - there is a free plan: https://brave.com/search/api/)
• Install the Fetch MCP server - [optional]

Start the Tiny Gordon Agent

• You can change of model by updating the .env file
• ✋ This demo is using tools, so my advice is to stay with ai/qwen2.5:latest

If you are on macOS

1. The first time you need to initialise the vector database with the documents (./data-loader/docs)

rm -rf data
docker compose --profile init up --build

2. Then, the next time, you can start it with:

docker compose up

If you are on Linux:

1. The first time you need to initialise the vector database with the documents (./data-loader/docs)

rm -rf data
docker compose --file compose.linux.yml --profile init up --build

2. Then, the next time, you can start it with:

docker compose --file compose.linux.yml up

If you are on Linux and DevContainer:

1. The first time you need to initialise the vector database with the documents (./data-loader/docs)

rm -rf data
docker compose --file compose.devcontainer.yml --profile init up --build

2. Then, the next time, you can start it with:

docker compose --file compose.devcontainer.yml up

🐙 Docker Compose will start 2 services:

• http://localhost:8000 to interact with Tiny Gordon using a web UI.
• http://localhost:9000/ to interact with Tiny Gordon using a REST API.

Chat with Tiny Gordon

Try the following sentences:

- what is your name?
- question: (it triggers the `question_about_something` tool to search the answer in its database)
- search with brave what is docker bake (it triggers the `brave_web_search` tool to search the answer on the web)
- ...

Using the API

Initialize a session:

curl -X POST http://localhost:9000/apps/tiny-gordon/users/bob/sessions/bob_session_42 \
  -H "Content-Type: application/json" \
  -d '{"state": {}}' | jq '.'

What is your name?:

curl -X POST http://localhost:9000/run \
-H "Content-Type: application/json" \
-d '{
    "appName": "tiny-gordon",
    "userId": "bob",
    "sessionId": "bob_session_42",
    "newMessage": {
        "role": "user",
        "parts": [{
            "text": "What is your name?"
        }]
    }
}' | jq '.'

Containers list:

curl -X POST http://localhost:9000/run \
-H "Content-Type: application/json" \
-d '{
    "appName": "tiny-gordon",
    "userId": "bob",
    "sessionId": "bob_session_42",
    "newMessage": {
        "role": "user",
        "parts": [{
            "text": "How to get the list of the running docker containers"
        }]
    }
}' | jq '.'

Development mode using DevContainer

👀 look at the .devcontainer/ directory

First time - Initialize the Python environment

Start DevContainer.

Create virtual environment:

python -m venv tmp

choose the name you want

Activate virtual environment:

source tmp/bin/activate

To deactivate virtual environment

deactivate

Add dependencies:

pip install -r requirements.txt

Start the agent

# activate the virtual environment: source tmp/bin/activate
adk web

Always use the python virtual environment when running the agent

In progress

• talk-to-moby: a MCP server to run docker commands