Skip to main content

​
Action Context

Use action_context to send actions with user identity: 
from value import initialize_sync

client = initialize_sync(agent_secret="your_agent_secret")

with client.action_context(user_id="user_123", anonymous_id="anon_456") as ctx:
    # Your agent logic here
    result = process_data()
    
    ctx.send(
        action_name="data_processed",
        **{"value.action.description": "Processed user data"}
    )

​
Context Parameters

  • user_id (optional): Identified user ID
  • anonymous_id (required): Anonymous session/request ID

​
Sending Actions

Within an action context, use ctx.send(): 
with client.action_context(user_id="user_123", anonymous_id="session_abc") as ctx:
    # Send an LLM call action
    ctx.send(
        action_name="llm_call",
        **{
            "value.action.description": "Generated response",
            "model": "gpt-4",
            "input_tokens": 1000,
            "output_tokens": 500,
        }
    )
    
    # Send a tool call action
    ctx.send(
        action_name="tool_call",
        **{
            "value.action.description": "Called external API",
            "tool_name": "stripe_api",
        }
    )

​
Standard Attributes

Use value.action.* prefix for standard attributes:
AttributeDescription
value.action.nameAction name (set automatically)
value.action.descriptionHuman-readable description
value.action.user_idUser ID (set from context)
value.action.anonymous_idAnonymous ID (set from context)
Custom attributes are stored in value.action.user_attributes.

​
Async Usage

import asyncio
from value import initialize_async

async def main():
    client = await initialize_async(agent_secret="your_agent_secret")
    
    with client.action_context(user_id="user_123", anonymous_id="anon_456") as ctx:
        result = await async_process()
        ctx.send(action_name="async_action")

asyncio.run(main())

​
Auto-Instrumentation

Automatically capture LLM calls without manual instrumentation: 
from value import initialize_sync, auto_instrument

client = initialize_sync(agent_secret="your_agent_secret")
auto_instrument(["gemini", "langchain"])

# LLM calls are now automatically traced
from google import genai
gemini_client = genai.Client(api_key="your-key")
response = gemini_client.models.generate_content(
    model="gemini-2.5-flash",
    contents=["Hello"]
)

​
Supported Libraries

LibraryExtraInstall
Google Generative AI (Gemini)genaipip install value-python[genai]
LangChainlangchainpip install value-python[langchain]

​
Disable Auto-Instrumentation

from value import uninstrument

uninstrument(["gemini"])  # Disable specific libraries
uninstrument()  # Disable all