TraceConfig

Configure trace behavior using TraceConfig objects across Python, JavaScript, and Java

TraceConfig

The TraceConfig object allows you to configure trace behavior programmatically in your application code. This provides an alternative to setting environment variables and allows for dynamic configuration.

Overview

TraceConfig helps you modify the observability level of your tracing. You can:

• Keep sensitive information from being logged for security reasons
• Limit the size of base64 encoded images to reduce payload size
• Control which parts of your LLM interactions are traced

Precedence Order

Configuration values are resolved in the following order:

1. Values set in the TraceConfig object in code (highest priority)
2. Environment variables
3. Default values (lowest priority)

Python

Import

from openinference.instrumentation import TraceConfig

Usage

Create a TraceConfig instance and pass it to your instrumentor’s instrument() method:

from openinference.instrumentation import TraceConfig
from openinference.instrumentation.openai import OpenAIInstrumentor

config = TraceConfig(
    hide_llm_invocation_parameters=True,
    hide_inputs=False,
    hide_outputs=False,
    hide_input_messages=False,
    hide_output_messages=False,
    hide_input_images=True,
    hide_input_text=False,
    hide_output_text=False,
    hide_embeddings_vectors=False,
    hide_embeddings_text=False,
    base64_image_max_length=16000,
    hide_prompts=False,
    hide_choices=False,
)

OpenAIInstrumentor().instrument(
    tracer_provider=tracer_provider,
    config=config,
)

Parameters

Parameter	Type	Default	Description
hide_llm_invocation_parameters	bool	False	Hides LLM invocation parameters
hide_inputs	bool	False	Hides input.value and all input messages
hide_outputs	bool	False	Hides output.value and all output messages
hide_input_messages	bool	False	Hides all input messages (independent of HIDE_INPUTS)
hide_output_messages	bool	False	Hides all output messages (independent of HIDE_OUTPUTS)
hide_input_images	bool	False	Hides images from input messages
hide_input_text	bool	False	Hides text from input messages
hide_output_text	bool	False	Hides text from output messages
hide_embedding_vectors	bool	False	Deprecated: Use hide_embeddings_vectors instead
hide_embeddings_vectors	bool	False	Replaces embedding vectors with "__REDACTED__"
hide_embeddings_text	bool	False	Replaces embedding text with "__REDACTED__"
hide_prompts	bool	False	Hides LLM prompts (completions API)
hide_choices	bool	False	Hides LLM choices (completions API outputs)
base64_image_max_length	int	32000	Limits characters of a base64 encoding of an image

Example: Hiding Sensitive Data

# Hide all PII but keep structure visible
config = TraceConfig(
    hide_input_text=True,
    hide_output_text=True,
    hide_embeddings_text=True,
)

JavaScript

Import

import { OpenAIInstrumentation } from "@arizeai/openinference-instrumentation-openai"

Usage

Create a trace config object and pass it to your instrumentation constructor:

import { OpenAIInstrumentation } from "@arizeai/openinference-instrumentation-openai"

const traceConfig = {
  hideInputs: true,
  hideOutputs: false,
  hideInputImages: true,
  base64ImageMaxLength: 16000,
}

const instrumentation = new OpenAIInstrumentation({ traceConfig })

Parameters

Parameter	Type	Default	Description
hideInputs	boolean	false	Hides input.value and all input messages
hideOutputs	boolean	false	Hides output.value and all output messages
hideInputMessages	boolean	false	Hides all input messages
hideOutputMessages	boolean	false	Hides all output messages
hideInputImages	boolean	false	Hides images from input messages
hideInputText	boolean	false	Hides text from input messages
hideOutputText	boolean	false	Hides text from output messages
hideEmbeddingVectors	boolean	false	Replaces embedding vectors with "__REDACTED__"
hidePrompts	boolean	false	Hides LLM prompts
base64ImageMaxLength	number	32000	Limits characters of a base64 encoding of an image

Example: Partial Configuration

// Only specify what you want to override
// Everything else falls back to environment variables, then defaults
const traceConfig = { hideInputs: true }

const instrumentation = new OpenAIInstrumentation({ traceConfig })

Java

Import

import com.arize.instrumentation.TraceConfig;

Usage

Use the builder pattern to create a TraceConfig instance:

import com.arize.instrumentation.TraceConfig;

TraceConfig config = TraceConfig.builder()
    .hideInputs(true)
    .hideOutputs(false)
    .hideInputImages(true)
    .hideInputText(false)
    .hideOutputText(false)
    .hideInputEmbeddings(false)
    .hideOutputEmbeddings(false)
    .hidePromptTemplate(false)
    .hidePromptTemplateVariables(false)
    .hideToolParameters(false)
    .base64ImageMaxLength("16000")
    .build();

Parameters

Parameter	Type	Default	Description
hideInputs	boolean	false	Hides input values
hideOutputs	boolean	false	Hides output values
hideInputMessages	boolean	false	Hides all input messages
hideOutputMessages	boolean	false	Hides all output messages
hideInputImages	boolean	false	Hides images from input messages
hideOutputImages	boolean	false	Hides images from output messages
hideInputText	boolean	false	Hides text from input messages
hideOutputText	boolean	false	Hides text from output messages
hideInputAudio	boolean	false	Hides audio from input messages
hideOutputAudio	boolean	false	Hides audio from output messages
hideInputEmbeddings	boolean	false	Hides input embeddings
hideOutputEmbeddings	boolean	false	Hides output embeddings
hidePromptTemplate	boolean	false	Hides prompt templates
hidePromptTemplateVariables	boolean	false	Hides prompt template variables
hidePromptTemplateVersion	boolean	false	Hides prompt template version
hideToolParameters	boolean	false	Hides tool parameters
base64ImageMaxLength	String	"unlimited"	Limits characters of a base64 encoding of an image

Example: Default Configuration

// Use all defaults
TraceConfig config = TraceConfig.getDefault();

Redacted Content

When content is hidden due to privacy configuration settings, the value "__REDACTED__" is used as a placeholder across all languages. This constant value allows consumers of the trace data to identify that content was intentionally hidden rather than missing or empty.

Next Steps

• See Environment Variables for alternative configuration methods
• See Privacy Controls for comprehensive privacy scenarios