revenium-griptape 0.1.0

Universal Revenium middleware for Griptape framework - transparent AI usage metering across all providers

Revenium Universal Griptape Driver

Add automatic AI usage metering and cost tracking to your Griptape applications with zero code changes. Works with any LLM provider that Griptape supports - simply wrap your existing driver and get detailed analytics via Revenium.

What This Does

This package provides a universal wrapper for any Griptape prompt driver that automatically:

• Meters all AI usage across any provider and sends data to Revenium for analytics
• Tracks costs and token usage with provider-specific optimizations
• Maintains full compatibility with existing Griptape code
• Supports rich metadata for detailed reporting and cost allocation
• Auto-detects providers and uses the best available middleware

Provider Support Architecture

Tier 1: Direct Native Support ⚡

Best performance with optimized provider-specific middleware

• OpenAI → revenium-middleware-openai
• Anthropic → revenium-middleware-anthropic
• Ollama → revenium-middleware-ollama

Tier 2: Universal LiteLLM Support 🌐

Supports 100+ providers through LiteLLM integration

• Google/Gemini → via revenium-middleware-litellm
• Cohere → via revenium-middleware-litellm
• Azure OpenAI → via revenium-middleware-litellm
• AWS Bedrock → via revenium-middleware-litellm
• LiteLLM Proxy → via revenium-middleware-litellm (centralized proxy server)
• And 100+ more → See LiteLLM docs

How It Works

1. Wrap any Griptape driver with ReveniumDriver
2. Add optional metadata for tracking
3. Your AI calls automatically get metered and reported to Revenium
4. View analytics and cost data in your Revenium dashboard

Installation

# Base package
pip install revenium-griptape

# Install middleware for your providers
pip install revenium-middleware-openai      # For OpenAI (Tier 1)
pip install revenium-middleware-anthropic   # For Anthropic (Tier 1)  
pip install revenium-middleware-ollama      # For Ollama (Tier 1)
pip install revenium-middleware-litellm     # For all other providers (Tier 2)

Environment Setup

Create a .env file in your project:

# Required: Your Revenium API key for metering
REVENIUM_METERING_API_KEY=your_revenium_api_key_here

# Optional: Custom Revenium endpoint (defaults to production)
REVENIUM_METERING_BASE_URL=https://api.revenium.io/meter

# Tier 1 Provider API Keys (Direct Support)
OPENAI_API_KEY=your_openai_api_key_here
ANTHROPIC_API_KEY=your_anthropic_api_key_here
# OLLAMA_HOST=http://localhost:11434  # For local Ollama

# Tier 2 Provider API Keys (via LiteLLM - as needed)
GEMINI_API_KEY=your_google_gemini_api_key_here
COHERE_API_KEY=your_cohere_api_key_here
AWS_ACCESS_KEY_ID=your_aws_access_key_here        # For Bedrock
AWS_SECRET_ACCESS_KEY=your_aws_secret_key_here    # For Bedrock
AZURE_API_KEY=your_azure_openai_api_key_here      # For Azure OpenAI

# LiteLLM Proxy Configuration (optional - for centralized proxy server)
LITELLM_PROXY_URL=http://localhost:4000/chat/completions
LITELLM_API_KEY=sk-1234

# See LiteLLM docs for other provider environment variables

Tier 2 Environment Variables: For Tier 2 providers, environment variable names follow LiteLLM conventions. Each provider has specific variable names (e.g., GEMINI_API_KEY for Google, COHERE_API_KEY for Cohere). See the LiteLLM provider documentation for the complete list.

Quick Start

Universal Driver (Recommended)

import os
from dotenv import load_dotenv

# ✅ CRITICAL: Load environment variables BEFORE importing
load_dotenv()

from griptape.structures import Agent
from griptape.drivers.prompt.openai_chat_prompt_driver import OpenAiChatPromptDriver

# Universal driver works with ANY Griptape prompt driver
from revenium_griptape import ReveniumDriver

# Method 1: Wrap an existing Griptape driver
base_driver = OpenAiChatPromptDriver(
    model="gpt-4o-mini",
    api_key=os.getenv("OPENAI_API_KEY")
)

driver = ReveniumDriver(
    base_driver=base_driver,
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "user_api_key",
                "value": "user_key_value"
            }
        },
        "task_type": "qa"
    }
)

# Method 2: Auto-detect from model name
driver = ReveniumDriver(
    model="gpt-4o-mini",  # Auto-detects OpenAI
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "user_api_key",
                "value": "user_key_value"
            }
        },
        "task_type": "qa"
    }
)

# Use exactly like any Griptape driver - metering is automatic!
agent = Agent(prompt_driver=driver)
result = agent.run("What is the capital of France?")
print(result.output.value)

Direct Driver Usage

from revenium_griptape import ReveniumOpenAiDriver, ReveniumAnthropicDriver

# OpenAI (Direct)
openai_driver = ReveniumOpenAiDriver(
    model="gpt-4o-mini",
    api_key=os.getenv("OPENAI_API_KEY"),
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "openai_api_key",
                "value": "openai_key_value"
            }
        }
    }
)

# Anthropic (Direct)
anthropic_driver = ReveniumAnthropicDriver(
    model="claude-3-haiku-20240307",
    api_key=os.getenv("ANTHROPIC_API_KEY"),
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "anthropic_api_key",
                "value": "anthropic_key_value"
            }
        }
    }
)

# LiteLLM for 100+ providers
from revenium_griptape import ReveniumLiteLLMDriver

# Google Gemini via LiteLLM
gemini_driver = ReveniumLiteLLMDriver(
    model="gemini-pro",
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "gemini_api_key",
                "value": "gemini_key_value"
            }
        }
    }
)

# LiteLLM Proxy
proxy_driver = ReveniumLiteLLMDriver(
    model="gpt-4",  # Model available through your proxy
    proxy_url=os.getenv("LITELLM_PROXY_URL"),
    proxy_api_key=os.getenv("LITELLM_API_KEY"),
    usage_metadata={
        "subscriber": {
            "id": "user-123",
            "email": "[e-mail]",
            "credential": {
                "name": "proxy_api_key",
                "value": "proxy_key_value"
            }
        }
    }
)

Supported Metadata Fields

Add any of these optional fields to track usage across different dimensions:

Field	Description	Example
trace_id	Session or conversation ID	"chat-session-123"
task_type	Type of AI task	"summarization", "qa"
subscriber	User information object	{"id": "user-456", "email": "[e-mail]"}
subscriber.id	User ID	"user-456"
subscriber.email	User email	"[e-mail]"
subscriber.credential	API credential info	{"name": "api_key", "value": "key_value"}
organization_id	Team or department	"sales-team"
subscription_id	Billing plan	"enterprise-plan"
product_id	Product or feature	"ai-assistant"
agent	AI agent identifier	"support-bot-v2"

Examples

View Complete Examples →

Example	Description
Universal Driver	Recommended - Multi-provider with auto-detection
OpenAI Direct	Direct OpenAI integration
Anthropic Direct	Direct Anthropic/Claude integration
LiteLLM Client	LiteLLM direct client integration
LiteLLM Proxy	LiteLLM proxy server integration

Each example is self-contained with environment setup, error handling, and detailed explanations.

Quick Start: Run python examples/universal_example.py after setting up your environment variables.