tokenator 0.1.15

Token usage tracking wrapper for LLMs

Tokenator : Track and analyze LLM token usage and cost

Have you ever wondered :

• How many tokens does your AI agent consume?
• How much does it cost to do run a complex AI workflow with multiple LLM providers?
• How much money/tokens did you spend today on developing with LLMs?

Afraid not, tokenator is here! With tokenator's easy to use API, you can start tracking LLM usage in a matter of minutes.

Get started with just 3 lines of code!

Installation

pip install tokenator

Usage

OpenAI

from openai import OpenAI
from tokenator import tokenator_openai

openai_client = OpenAI(api_key="your-api-key")

# Wrap it with Tokenator
client = tokenator_openai(openai_client)

# Use it exactly like the OpenAI client
response = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Hello!"}]
)

Works with AsyncOpenAI and streaming=True as well! Note : When streaming, don't forget to add stream_options={"include_usage": True} to the create() call!

Cost Analysis

from tokenator import usage

# Get usage for different time periods
usage.last_hour()
usage.last_day()
usage.last_week()
usage.last_month()

# Custom date range
usage.between("2024-03-01", "2024-03-15")

# Get usage for different LLM providers
usage.last_day("openai")
usage.last_day("anthropic")
usage.last_day("google")

Example usage object

print(cost.last_hour().model_dump_json(indent=4))

{
    "total_cost": 0.0004,
    "total_tokens": 79,
    "prompt_tokens": 52,
    "completion_tokens": 27,
    "providers": [
        {
            "total_cost": 0.0004,
            "total_tokens": 79,
            "prompt_tokens": 52,
            "completion_tokens": 27,
            "provider": "openai",
            "models": [
                {
                    "total_cost": 0.0004,
                    "total_tokens": 79,
                    "prompt_tokens": 52,
                    "completion_tokens": 27,
                    "model": "gpt-4o-2024-08-06"
                }
            ]
        }
    ]
}

Features

• Drop-in replacement for OpenAI, Anthropic client
• Automatic token usage tracking
• Cost analysis for different time periods
• SQLite storage with zero configuration
• Thread-safe operations
• Minimal memory footprint
• Minimal latency footprint

Anthropic

from anthropic import Anthropic, AsyncAnthropic
from tokenator import tokenator_anthropic

anthropic_client = AsyncAnthropic(api_key="your-api-key")

# Wrap it with Tokenator
client = tokenator_anthropic(anthropic_client)

# Use it exactly like the Anthropic client
response = await client.messages.create(
    model="claude-3-5-haiku-20241022",
    messages=[{"role": "user", "content": "hello how are you"}],
    max_tokens=20,
)

print(response)

print(usage.last_execution().model_dump_json(indent=4))
"""
{
    "total_cost": 0.0001,
    "total_tokens": 23,
    "prompt_tokens": 10,
    "completion_tokens": 13,
    "providers": [
        {
            "total_cost": 0.0001,
            "total_tokens": 23,
            "prompt_tokens": 10,
            "completion_tokens": 13,
            "provider": "anthropic",
            "models": [
                {
                    "total_cost": 0.0004,
                    "total_tokens": 79,
                    "prompt_tokens": 52,
                    "completion_tokens": 27,
                    "model": "claude-3-5-haiku-20241022"
                }
            ]
        }
    ]
}
"""

xAI

You can use xAI models through the openai SDK and track usage using provider parameter in tokenator.

from openai import OpenAI
from tokenator import tokenator_openai

xai_client = OpenAI(
            api_key=os.getenv("XAI_API_KEY"),
            base_url="https://api.x.ai/v1"
        )

# Wrap it with Tokenator
client = tokenator_openai(xai_client, db_path=temp_db, provider="xai")

# Use it exactly like the OpenAI client but with xAI models
response = client.chat.completions.create(
    model="grok-2-latest",
    messages=[{"role": "user", "content": "Hello!"}]
)

print(response)

print(usage.last_execution())

Other AI model providers through openai SDKs

Today, a variety of AI companies have made their APIs compatible to the openai SDK. You can track usage of any such AI models using tokenator's provider parameter.

For example, let's see how we can track usage of perplexity tokens.

from openai import OpenAI
from tokenator import tokenator_openai

perplexity_client = OpenAI(
            api_key=os.getenv("PERPLEXITY_API_KEY"),
            base_url="https://api.perplexity.ai"
        )

# Wrap it with Tokenator
client = tokenator_openai(perplexity_client, db_path=temp_db, provider="perplexity")

# Use it exactly like the OpenAI client but with perplexity models
response = client.chat.completions.create(
    model="llama-3.1-sonar-small-128k-online",
    messages=[{"role": "user", "content": "Hello!"}]
)

print(response)

print(usage.last_execution())

print(usage.provider("perplexity"))

---

Most importantly, none of your data is ever sent to any server.

License

MIT