definable 0.2.4

Infrastructure for building and deploying AI agents

Definable

Infrastructure for building and deploying AI agents with a simple YAML configuration and base class extension pattern.

Features

• Simple Base Class: Extend AgentBox to create your agent
• YAML Configuration: Configure builds with declarative YAML files
• Docker Packaging: Build Docker images with a single command
• FastAPI Integration: Automatic REST API generation with OpenAPI docs
• CLI Tools: Complete toolkit with init, build, serve, push, and config commands
• Project Scaffolding: Generate new agent projects with templates
• Remote Deployment: Push agents to deployment servers
• Hook System: Pre and post-processing hooks for middleware

Quick Start

1. Install Definable

pip install definable

2. Initialize a New Project

# Create a new agent project with interactive prompts
definable init

# Or specify details directly
definable init --name my-agent --description "My custom agent" --version 1.0.0

This creates:

• main.py - Your agent implementation
• agent.yaml - Configuration file
• .agentignore - Files to exclude from builds
• README.md - Project documentation

3. Create Your Agent

The generated main.py provides a template, or create your own:

# main.py
from definable import AgentBox, AgentInput, AgentOutput, AgentInfo
from pydantic import Field

class SampleAgentInput(AgentInput):
    message: str = Field(description="Input message to process")

class SampleAgentOutput(AgentOutput):
    response_message: str = Field(description="Processed response message")

class DemoAgent(AgentBox):
    def setup(self):
        """Initialize your agent - called once on startup"""
        self.name = 'demo-agent'
        self.version = '1.0.0'
        print("Demo agent initialized!")
    
    def invoke(self, agent_input: SampleAgentInput) -> SampleAgentOutput:
        """Main agent logic - called for each request"""
        processed_message = f"Processed: {agent_input.message.upper()}"
        return SampleAgentOutput(response_message=processed_message)
    
    def info(self) -> AgentInfo:
        """Return agent metadata for API documentation"""
        return AgentInfo(
            name=self.name,
            description="A simple demo agent that processes messages",
            version=self.version,
            input_schema=SampleAgentInput.model_json_schema(),
            output_schema=SampleAgentOutput.model_json_schema()
        )

4. Configure Your Agent

The agent.yaml file configures the build and deployment:

# agent.yaml
build:
  python_version: "3.11"
  dependencies:
    - "requests>=2.28.0"
    - "openai"  # Add your specific dependencies
  system_packages:
    - "curl"
  environment_variables:
    - API_KEY  # Environment variables to pass through

agent: "main.py:DemoAgent"

platform:
  name: "demo-agent"
  description: "A simple demo agent for testing"
  version: "1.0.0"

concurrency:
  max_concurrent_requests: 50
  request_timeout: 300

5. Develop and Deploy

# Serve locally for development (includes hot reload)
definable serve -p 8000

# Build Docker image for production
definable build -t my-agent

# Push to deployment server (if configured)
definable push

# Manage configuration
definable config

CLI Commands

• definable init - Initialize a new agent project with templates
• definable serve - Serve agent locally for development
• definable build - Build Docker image from agent configuration
• definable push - Deploy agent to remote deployment server
• definable config - Manage global configuration settings

CLI Options

# Serve with custom options
definable serve --port 3000 --host localhost --file custom-agent.yaml

# Build with custom tag and config
definable build -t my-agent:v1.2.0 -f production-agent.yaml

# Push with deployment options
definable push -n my-agent

Advanced Features

Hook System

Add pre and post-processing logic:

class MyAgent(AgentBox):
    def pre_hook(self):
        """Called before invoke()"""
        print("Pre-processing...")
    
    def post_hook(self):
        """Called after invoke()"""
        print("Post-processing...")

Flexible Input/Output

You can use any types for inputs and outputs:

def invoke(self, agent_input: dict) -> str:
    """Simple string-to-string agent"""
    return f"Echo: {agent_input.get('message', '')}"

Environment Configuration

The framework supports environment variables and secrets:

build:
  environment_variables:
    - DATABASE_URL
    - API_SECRET
    - DEBUG=false

API Documentation

When served, your agent automatically gets:

• OpenAPI/Swagger docs at /docs
• Health check endpoint at /health
• Agent info endpoint at /info

Requirements

• Python: 3.9 or higher
• Docker: Required for building images
• Dependencies: FastAPI, Uvicorn, Pydantic, Click, Docker SDK

License

MIT License - see LICENSE file for details.