DB-ADK: Database-Driven Agent Development Kit

DB-ADK is a Python package that extends Google's Agent Development Kit (ADK) to create a fully database-driven approach for building and managing agent networks. This approach minimizes explicit Python coding by storing agent configurations, prompts, and tool definitions in a SQLite database.

Key Features

• Database-Driven: Store all agent and tool configurations in SQLite
• Minimal Python Coding: Define new agents and tools through the database or API without writing new code
• Dynamic Loading: Automatically load tools and create agents from database records
• Agent Networks: Build complex multi-agent systems with different relationship types
• API Interfaces: REST API and CLI for managing agents, tools, and networks
• Import/Export: Export and import agents in JSON or YAML format for easy sharing and backup
• Extensible: Easily add new tools and agent types

Installation

Using uv (Recommended)

# Clone the repository
git clone https://github.com/yourusername/db-adk.git
cd db-adk

# Install using uv
uv install -e .

Using pip

# Clone the repository
git clone https://github.com/yourusername/db-adk.git
cd db-adk

# Install the package
pip install -e .

Requirements

• Python 3.9+
• SQLite (built into Python)
• Google ADK (google-adk)

Quick Start

1. Set up environment variables

Create a .env file with your ADK configuration:

# SQLite database path (default: db_adk.sqlite in the current directory)
DB_PATH=db_adk.sqlite
ADK_API_KEY=your_api_key
DEFAULT_MODEL=gemini-1.5-pro

2. Initialize the database

# Initialize the database
db-adk db init

3. Create agents and tools

You can create agents and tools using the CLI, REST API, or by directly inserting records into the database.

Using the CLI:

# Create a tool
db-adk tools create \
  --name "get_weather" \
  --description "Get weather information for a location" \
  --type "python_function" \
  --module "my_tools.weather" \
  --function "get_weather" \
  --schema weather_schema.json

# Create an agent
db-adk agents create \
  --name "weather_assistant" \
  --description "Provides weather information" \
  --type "LLM" \
  --prompt "You are a weather assistant that provides weather information." \
  --model "gemini-1.5-pro"

# Assign tool to agent
db-adk agent-tools assign --agent-id 1 --tool-id 1

4. Run an agent

# Create input.json with the query
echo '{"query": "What is the weather in New York?"}' > input.json

# Run the agent
db-adk agents run 1 --input input.json

5. Run the REST API server

# Start the server
db-adk serve

6. Import/Export Agents

DB-ADK supports exporting and importing agents in both JSON and YAML formats:

# Export an agent to JSON (default)
db-adk agents export 1 --output weather_agent.json

# Export an agent to YAML
db-adk agents export 1 --output weather_agent.yaml --format yaml

# Export all agents to a directory
db-adk agents export-all --output-dir agents_backup --format yaml

# Import an agent from file (format auto-detected)
db-adk agents import --file weather_agent.json

# Import all agents from a directory
db-adk agents import-dir --directory agents_backup

Architecture

DB-ADK consists of the following components:

1. Database Layer: SQLAlchemy models and repositories for database interactions
2. Core Layer: Dynamic agent and tool factories
3. API Layer: REST API and CLI interfaces
4. Import/Export: JSON and YAML import/export functionality
5. Utils: Logging and schema validation utilities

Database Schema

The database schema includes the following tables:

• agents: Agent definitions (configurations, prompts, descriptions)
• tools: Tool definitions (functions, parameters, descriptions)
• agent_tools: Mapping between agents and tools
• agent_relationships: Relationships between agents (for multi-agent systems)

Examples

Multi-Agent Network Example

The package includes an example of a travel planning system with a coordinator agent and specialized agents for weather information and itinerary planning:

# Run the multi-agent network example
python -m db_adk.examples.multi_agent_network

This example demonstrates:

• Creating a coordinator agent
• Creating specialized agents
• Assigning tools to agents
• Setting up relationships between agents
• Running a multi-agent network

API Reference

REST API Endpoints

• /agents/: CRUD operations for agents
• /tools/: CRUD operations for tools
• /agent-tools/: Manage agent-tool mappings
• /agent-relationships/: Manage agent relationships
• /agents/{agent_id}/run: Run an agent
• /networks/{coordinator_id}/run: Run an agent network
• /agents/{agent_id}/export: Export an agent
• /agents/export-all: Export all agents
• /agents/import: Import an agent

CLI Commands

• db-adk db init: Initialize the database
• db-adk agents list: List all agents
• db-adk agents create: Create a new agent
• db-adk agents run: Run an agent
• db-adk agents export: Export an agent to JSON or YAML
• db-adk agents export-all: Export all agents to a directory
• db-adk agents import: Import an agent from a file
• db-adk agents import-dir: Import agents from a directory
• db-adk tools list: List all tools
• db-adk tools create: Create a new tool
• db-adk agent-tools assign: Assign a tool to an agent
• db-adk agent-tools list: List agent-tool mappings
• db-adk relationships create: Create a relationship between agents
• db-adk networks run: Run an agent network
• db-adk serve: Start the REST API server

Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

License

This project is licensed under the MIT License - see the LICENSE file for details.