tokentally 0.4.0

Track AI API usage with TokenTally

TokenTally Python SDK

Track your AI API usage with TokenTally. Get real-time analytics, cost tracking, and usage insights for Claude, OpenAI, and other AI providers.

Installation

pip install tokentally

Quick Start

from tokentally import TokenTally

# Initialize the client with your API key
tt = TokenTally(api_key="tt_your_api_key")

# Track usage manually
response = tt.track(
    tokens_in=150,
    tokens_out=300,
    model="claude-3-sonnet-20240229",
    provider="anthropic",
    metadata={
        "feature": "chat",
        "user_id": "user123",
    }
)

print(f"Recorded! Cost: ${response.cost_usd:.6f}")

Usage with Context Manager

The context manager automatically tracks timing:

from tokentally import TokenTally
import anthropic

tt = TokenTally(api_key="tt_your_api_key")
client = anthropic.Anthropic()

with tt.track_usage(
    model="claude-3-sonnet-20240229",
    metadata={"feature": "summarization"}
) as ctx:
    # Make your API call
    response = client.messages.create(
        model="claude-3-sonnet-20240229",
        max_tokens=1000,
        messages=[{"role": "user", "content": "Summarize this..."}]
    )

    # Set the usage from the response
    ctx.set_usage(
        tokens_in=response.usage.input_tokens,
        tokens_out=response.usage.output_tokens,
        stop_reason=response.stop_reason,
    )

# Usage is automatically sent when the context exits
print(f"Cost: ${ctx.response.cost_usd:.6f}")

Configuration

tt = TokenTally(
    api_key="tt_your_api_key",
    base_url="https://api.tokentally.cloud",  # Optional custom URL
    timeout=30.0,  # Request timeout in seconds
)

Error Handling

from tokentally import TokenTally, RateLimitError, AuthenticationError, TokenTallyError

tt = TokenTally(api_key="tt_your_api_key")

try:
    tt.track(tokens_in=100, tokens_out=200, model="claude-3-sonnet")
except RateLimitError:
    print("Rate limit exceeded. Upgrade your plan for unlimited usage.")
except AuthenticationError:
    print("Invalid API key. Check your credentials.")
except TokenTallyError as e:
    print(f"Error tracking usage: {e}")

Metadata

Add custom metadata to categorize your usage. The application and feature keys are special -- they appear as filterable dropdowns in the TokenTally dashboard, making it easy to break down usage by app or feature.

tt.track(
    tokens_in=100,
    tokens_out=200,
    model="claude-3-sonnet-20240229",
    metadata={
        "application": "my-app",        # Filterable in dashboard
        "feature": "chat_completion",    # Filterable in dashboard
        "user_id": "user123",
        "environment": "production",
        "request_id": "abc-123",
    }
)

You can add any other key-value pairs to metadata for your own tracking purposes.

Image Generation

Track image generation APIs like Leonardo, DALL-E, and Midjourney using credits, resolution, and quality parameters:

# Leonardo AI
tt.track(
    model="leonardo-phoenix",
    provider="leonardo",
    credits=15,
    resolution="1024x1024",
    quality="high",
)

# OpenAI DALL-E
tt.track(
    model="dall-e-3",
    provider="openai",
    resolution="1024x1024",
    quality="hd",
)

# With context manager
with tt.track_usage(model="leonardo-phoenix", provider="leonardo") as ctx:
    response = leonardo.generate(prompt="A sunset over mountains")

    ctx.set_usage(
        credits=response.credits_used,
        resolution="1024x1024",
        quality="high",
    )

License

MIT License - see LICENSE for details.