value-python 0.1.6

Python SDK to track AI agents with Value actions and auto-instrument LLM calls (Gemini, LangChain)

Value Python SDK

Python SDK to track AI agents with Value actions and auto-instrument LLM calls (Gemini, LangChain).

Features

• Value Actions: Track agent behavior using action_context with user_id and anonymous_id, send custom actions via ctx.send()
• Auto-Instrumentation: Automatically capture LLM calls from Gemini and LangChain with zero code changes
• OpenTelemetry-Based: Built on OpenTelemetry for standardized, vendor-neutral observability

Installation

Basic Installation

Install the core SDK without auto-instrumentation dependencies:

pip install value-python

With Google GenAI Auto-Instrumentation

Install with Google Generative AI (Gemini) auto-instrumentation support:

pip install value-python[genai]

With LangChain Auto-Instrumentation

Install with LangChain auto-instrumentation support:

pip install value-python[langchain]

With All Auto-Instrumentation Libraries

Install with all supported auto-instrumentation libraries:

pip install value-python[all]

Multiple Extras

You can also install multiple extras:

pip install value-python[genai,langchain]

Supported Platforms

• Python: 3.9, 3.10, 3.11, 3.12, 3.13
• Operating Systems: Linux, macOS, Windows

Quick Start

Basic Usage

import asyncio
from value import initialize_async

async def main():
    # agent_secret is required
    client = await initialize_async(agent_secret="your-agent-secret")

    async def process_data(data: str) -> str:
        print(f"Processing data: {data}")
        await asyncio.sleep(0.5)
        result = data.upper()

        with client.action_context(user_id="user123", anonymous_id="anon456") as ctx:
            ctx.send(
                action_name="transform_data",
                **{"value.action.description": f"Transformed data from {len(data)} to {len(result)} characters"}
            )
        return result

    result = await process_data("hello async world")
    print(f"Result: {result}")

asyncio.run(main())

Synchronous Usage

from value import initialize_sync

# agent_secret is required
client = initialize_sync(agent_secret="your-agent-secret")

with client.action_context(user_id="user123", anonymous_id="anon456") as ctx:
    # Your code here
    ctx.send(action_name="my_action", **{"custom.attribute": "value"})

Auto-Instrumentation

Enable automatic tracing for supported AI libraries:

from value import initialize_sync, auto_instrument

# Initialize the client with agent_secret
client = initialize_sync(agent_secret="your-agent-secret")

# Auto-instrument specific libraries
auto_instrument(["gemini", "langchain"])

# Or auto-instrument all available libraries
auto_instrument()

Google GenAI Example

from value import initialize_sync, auto_instrument
from google import genai

# Initialize Value client with agent_secret and auto-instrument
client = initialize_sync(agent_secret="your-agent-secret")
auto_instrument(["gemini"])

# Use Gemini as usual - traces are automatically captured
gemini_client = genai.Client(api_key="your-api-key")
response = gemini_client.models.generate_content(
    model="gemini-2.5-flash",
    contents=["Write a poem about tracing"]
)

print(response.text)

Configuration

The agent_secret is passed directly to initialize_sync() or initialize_async(). Additional configuration can be set using environment variables:

Variable	Description	Default
VALUE_OTEL_ENDPOINT	OpenTelemetry collector endpoint	http://localhost:4317
VALUE_BACKEND_URL	Value Control Plane backend URL	Required
VALUE_SERVICE_NAME	Service name for OpenTelemetry resource	value-control-agent
VALUE_CONSOLE_EXPORT	Enable console span exporter for debugging	false

Supported Auto-Instrumentation Libraries

Library	Extra	Instrumentor
Google Generative AI (Gemini)	genai	opentelemetry-instrumentation-google-generativeai
LangChain	langchain	opentelemetry-instrumentation-langchain

API Reference

Core Functions

• initialize_sync(agent_secret) - Initialize a synchronous Value client
• initialize_async(agent_secret) - Initialize an asynchronous Value client
• auto_instrument(libraries=None) - Enable auto-instrumentation for specified libraries
• uninstrument(libraries=None) - Disable auto-instrumentation
• get_supported_libraries() - Get list of supported library names
• is_library_available(library) - Check if a library's instrumentation is installed

Client Methods

• action_context(user_id=None, anonymous_id=None) - Create a context for sending actions
• ctx.send(action_name, **attributes) - Send an action with custom attributes

Development

Setup

# Clone the repository
git clone https://github.com/valmi-io/value-python.git
cd value-python

# Install dependencies
poetry install

# Install with all extras for development
poetry install --extras all

Running Tests

# Run tests
poetry run pytest

# Run tests with coverage
poetry run pytest --cov=value --cov-report=html

# Run specific test file
poetry run pytest tests/test_client.py

Code Quality

# Format code
poetry run black src/ tests/

# Lint code
poetry run ruff check src/ tests/

# Type check
poetry run mypy src/

Publishing

The package is automatically published to PyPI when a new release is created on GitHub.

Manual Publishing

# Build the package
poetry build

# Publish to TestPyPI (for testing)
poetry publish -r testpypi

# Publish to PyPI
poetry publish

License

MIT License - see LICENSE for details.

Contributing

Contributions are welcome! Please read our Contributing Guide for details.

Links

• Documentation
• PyPI Package
• Issue Tracker