demiurg 0.1.16

AI Agent Framework for building intelligent agents with multiple LLM providers

Demiurg SDK

A powerful AI agent framework for building production-ready conversational agents with support for multiple LLM providers and external tool integrations.

Features

• 🚀 Clean API - Simple, intuitive agent initialization
• 🔌 Multi-Provider Support - OpenAI with more providers coming soon
• 💰 Flexible Billing - Choose who pays for API calls (builder or end-user)
• 🛠️ Composio Integration - Connect to 150+ external services with OAuth
• 📬 Built-in Messaging - Queue management and conversation history
• 📁 Multimodal Support - Handle images, audio, text, and files
• 🎨 OpenAI Tools - Image generation (DALL-E 3), TTS, transcription
• ⚡ Progress Indicators - Real-time feedback for long operations
• 🏗️ Production Ready - Error handling, logging, and scalability

Installation

pip install demiurg

Quick Start

Simple Agent

from demiurg import Agent, OpenAIProvider

# Create an agent with OpenAI
agent = Agent(OpenAIProvider())

# Or with user-based billing
agent = Agent(OpenAIProvider(), billing="user")

Agent with External Tools (Composio)

from demiurg import Agent, OpenAIProvider, Composio

# Create agent with Twitter and GitHub access
agent = Agent(
    OpenAIProvider(),
    Composio("TWITTER", "GITHUB"),
    billing="user"
)

Custom Configuration

from demiurg import Agent, OpenAIProvider, Config

config = Config(
    name="My Assistant",
    description="A helpful AI assistant",
    model="gpt-4o",
    temperature=0.7,
    show_progress_indicators=True
)

agent = Agent(OpenAIProvider(), config=config)

Core Concepts

Billing Modes

The SDK supports two billing modes:

• "builder" (default) - API calls are charged to the agent builder's account
• "user" - API calls are charged to the end user's account

# Builder pays for all API calls
agent = Agent(OpenAIProvider(), billing="builder")

# End users pay for their own API calls
agent = Agent(OpenAIProvider(), billing="user")

Composio Integration

Connect your agents to external services like Twitter, GitHub, Gmail, and 150+ more:

# Configure Composio tools
agent = Agent(
    OpenAIProvider(),
    Composio("TWITTER", "GITHUB", "GMAIL"),
    billing="user"
)

# Check if user has connected their account
status = await agent.check_composio_connection("TWITTER", user_id)

# Handle OAuth flow in conversation
if not status["connected"]:
    await agent.handle_composio_auth_in_conversation(message, "TWITTER")

Create a composio-tools.txt file in your project root:

TWITTER=ac_your_twitter_config_id
GITHUB=ac_your_github_config_id
GMAIL=ac_your_gmail_config_id

Progress Indicators

Long operations automatically show progress messages:

config = Config(show_progress_indicators=True)  # Enabled by default

# Users will see:
# "🎨 Creating your image... This may take a moment."
# "🎵 Transcribing audio... This may take a moment."

Message Handling

Sending Messages

from demiurg import send_text, send_file

# Send text message
await send_text(conversation_id, "Hello from my agent!")

# Send file with caption
await send_file(
    conversation_id, 
    "/path/to/image.png", 
    caption="Here's your generated image!"
)

Processing Messages

from demiurg import Message

# Process user message
message = Message(
    content="Generate an image of a sunset",
    user_id="user123",
    conversation_id="conv456"
)

response = await agent.process_message(message)

Conversation History

from demiurg import get_conversation_history

# Get formatted history for LLM context
messages = await get_conversation_history(
    conversation_id,
    limit=50,
    provider="openai"  # Formats for specific provider
)

Built-in OpenAI Tools

When using OpenAI provider with tools enabled:

config = Config(use_tools=True)
agent = Agent(OpenAIProvider(), config=config)

Available tools:

• generate_image - Create images with DALL-E 3
• text_to_speech - Convert text to natural speech
• transcribe_audio - Transcribe audio files

Custom Agents

Basic Custom Agent

from demiurg import Agent, OpenAIProvider, Message

class MyCustomAgent(Agent):
    def __init__(self):
        super().__init__(
            OpenAIProvider(),
            billing="user"
        )
    
    async def process_message(self, message: Message, content=None) -> str:
        # Add custom preprocessing
        if "urgent" in message.content.lower():
            return await self.handle_urgent_request(message)
        
        # Use standard processing
        return await super().process_message(message, content)

Agent with Custom Tools

class ToolAgent(Agent):
    def __init__(self):
        config = Config(use_tools=True)
        super().__init__(OpenAIProvider(), config=config)
        
        # Register custom tool
        self.register_tool(
            {
                "type": "function",
                "function": {
                    "name": "get_weather",
                    "description": "Get current weather",
                    "parameters": {
                        "type": "object",
                        "properties": {
                            "location": {"type": "string"}
                        },
                        "required": ["location"]
                    }
                }
            },
            self.get_weather
        )
    
    async def get_weather(self, location: str) -> str:
        # Implement weather fetching
        return f"Weather in {location}: Sunny, 72°F"

File Handling

The SDK automatically handles various file types:

# Images are analyzed with vision models
# Audio files are automatically transcribed
# Text files have their content extracted

# File size limit: 10MB
# Supported image formats: PNG, JPEG, WEBP, GIF
# Supported audio formats: MP3, WAV, M4A, and more

Error Handling

from demiurg.exceptions import (
    DemiurgError,      # Base exception
    ConfigurationError,# Configuration issues
    MessagingError,    # Messaging failures
    ProviderError,     # LLM provider errors
    FileError,         # File operation failures
    ToolError          # Tool execution errors
)

try:
    response = await agent.process_message(message)
except ProviderError as e:
    # Handle LLM provider issues
    logger.error(f"Provider error: {e}")
except DemiurgError as e:
    # Handle other Demiurg errors
    logger.error(f"Agent error: {e}")

Environment Variables

Required environment variables:

# Core Configuration
DEMIURG_BACKEND_URL=http://backend:3000  # Backend API URL
DEMIURG_AGENT_TOKEN=your_token          # Authentication token
DEMIURG_AGENT_ID=your_agent_id          # Unique agent identifier

# Provider Keys
OPENAI_API_KEY=your_openai_key          # For OpenAI provider

# Composio Integration (optional)
COMPOSIO_API_KEY=your_composio_key      # For external tools
COMPOSIO_TOOLS=TWITTER,GITHUB,GMAIL    # Comma-separated toolkits

# Advanced Settings
DEMIURG_USER_ID=builder_user_id        # Builder's user ID (for billing)
TOOL_PROVIDER=composio                  # Tool provider selection

Advanced Features

Message Queue System

The SDK includes automatic message queuing to prevent race conditions:

# Messages are automatically queued per conversation
# Prevents issues when multiple messages arrive simultaneously
# No additional configuration needed - it just works!

Multimodal Capabilities

# Process images with vision models
if message contains image:
    # Automatically analyzed with GPT-4V
    
# Handle audio messages
if message contains audio:
    # Automatically transcribed with Whisper
    
# Text file processing
if message contains text file:
    # Content extracted and provided to LLM

Production Deployment

# Health check endpoint
@app.get("/health")
async def health_check():
    return await agent.health_check()

# Queue status monitoring
@app.get("/queue-status")
async def queue_status():
    return await agent.get_queue_status()

Architecture

The SDK follows a modular architecture:

• Agent: Core class that orchestrates everything
• Providers: LLM integrations (OpenAI, etc.)
• Tools: External service integrations (Composio)
• Messaging: Communication with Demiurg platform
• Utils: File handling, audio processing, etc.

Best Practices

1. Always use async/await - The SDK is built for async operations
2. Handle errors gracefully - Use try/except blocks with specific exceptions
3. Configure billing appropriately - Choose who pays for API calls
4. Set up Composio auth configs - Store in composio-tools.txt
5. Enable progress indicators - Better UX for long operations
6. Use appropriate models - GPT-4o for complex tasks, GPT-3.5 for simple ones

Migration Guide

From v0.1.10 to v0.1.11

# Old way
from demiurg import Agent, Config

config = Config(name="My Agent")
agent = Agent(config)

# New way (backward compatible)
from demiurg import Agent, OpenAIProvider

agent = Agent(OpenAIProvider())

Support

• Documentation: https://docs.demiurg.ai
• GitHub Issues: https://github.com/demiurg-ai/demiurg-sdk/issues
• Email: [e-mail]

License

Copyright © 2024 Demiurg AI. All rights reserved.

This is proprietary software. See LICENSE file for details.