Google ADK Multi Agent Ollama Application

Still this is the only Google ADK Ollama multi Agent application to run locally. It is not completed yet due to error from ADK Litellm and other function to communicate with Ollama and ADK.

image of krishigpt chatbot

Prerequisites

• Python 3.11 or higher
• pip (Python package installer)
• A Google Cloud Platform (GCP) account (for Gemini API access) will be good

Installation

1. Fork and Clone the Repository (if applicable):

   git clone <your_repository_url>
   cd <your_application_directory>

2. Create an .env file:

   Create a file named .env from env_copy in the root directory of your project. This file will store your API keys and other sensitive information.

   GOOGLE_API_KEY=<your_google_api_key>
   LANGCHAIN_API_KEY=<your_langsmith_api_key>
   LANGCHAIN_TRACING_V2="true"
   LANGCHAIN_PROJECT="Your_LangGraph_Project_Name"
   Antyhing else api key etc
   WEATHER_API_KEY=your-key
   USER_AGENT=MyLangGraphBot/1.0
   # MODEL=gemma3:4b
   MODEL=llama3.1:8b
   # BASE_URL=http://ollama_server_lite:11434
   BASE_URL=http://ollama_server:11434
   REMOTE_BASE_URL=https://models.github.ai/inference
   REMOTE_MODEL=openai/o4-mini
   

Obtaining API Keys

1. Google Gemini API Key [optional]

1. Go to Google AI Studio:
   • Google AI Studio
2. Follow the instructions to create a project and obtain an API key.
3. Alternatively, you can obtain a Google Cloud API key from the google cloud console.
   • Google Cloud Platform
4. Enable the Gemini API for your project.
5. Create API credentials.
6. Add the key to your .env file as GOOGLE_API_KEY.

Running the Application

1. Navigate to your application directory:

   cd <your_application_directory>
   docker-compose up --build

2. Run the application using the ADK CLI [optional]:

   adk web

Google ADK Local Development Notes.

• https://developers.googleblog.com/en/agent-development-kit-easy-to-build-multi-agent-applications

• 🐳 Running with Docker Compose You can also run the application using Docker Compose, which will spin up both the LangGraph agent and Ollama server.

• Start the services docker-compose up --build

• To stop ctrl + c

and

docker-compose down --remove-orphans

Always check the Docker images and containers. Keep your system and dangling images removed otherwise you system can freeze.

• This will launch:

🚀 agent_service: Your LangGraph agent on http://localhost:5000

🧠 ollama_server: The local model server running on http://localhost:11434

Use this for full isolation and easy multi-service orchestration. Recommended system at least 16 GB or 32 GB RAM, optional GPU, i7 or similar. As LLM will be downloaded into your docker container and it require around 5 GB. Higher system will run faster otherwise slow for response to keep patience.