Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/Arize-ai/openinference/llms.txt

Use this file to discover all available pages before exploring further.

​
Environment Variables

OpenInference supports configuration through environment variables, providing a convenient way to control tracing behavior without modifying your code.

​
Overview

Environment variables allow you to:
  • Configure tracing behavior globally across your application
  • Adjust observability levels between different environments (dev, staging, production)
  • Control privacy and security settings without code changes

​
Configuration Precedence

When both environment variables and TraceConfig are used:
  1. Values set in TraceConfig objects (highest priority)
  2. Environment variables
  3. Default values (lowest priority)

​
Supported Variables

​
Input/Output Controls

VariableTypeDefaultDescription
OPENINFERENCE_HIDE_INPUTSboolFalseHides input.value and all input messages. Input messages are hidden if either HIDE_INPUTS OR HIDE_INPUT_MESSAGES is true
OPENINFERENCE_HIDE_OUTPUTSboolFalseHides output.value and all output messages. Output messages are hidden if either HIDE_OUTPUTS OR HIDE_OUTPUT_MESSAGES is true
OPENINFERENCE_HIDE_INPUT_MESSAGESboolFalseHides all input messages (independent of HIDE_INPUTS)
OPENINFERENCE_HIDE_OUTPUT_MESSAGESboolFalseHides all output messages (independent of HIDE_OUTPUTS)

​
Message Content Controls

VariableTypeDefaultDescription
OPENINFERENCE_HIDE_INPUT_IMAGESboolFalseHides images from input messages (only applies when input messages are not already hidden)
OPENINFERENCE_HIDE_INPUT_TEXTboolFalseHides text from input messages (only applies when input messages are not already hidden)
OPENINFERENCE_HIDE_OUTPUT_TEXTboolFalseHides text from output messages (only applies when output messages are not already hidden)

​
LLM-Specific Controls

VariableTypeDefaultDescription
OPENINFERENCE_HIDE_LLM_INVOCATION_PARAMETERSboolFalseHides LLM invocation parameters (independent of input/output hiding)
OPENINFERENCE_HIDE_PROMPTSboolFalseHides LLM prompts (completions API)
OPENINFERENCE_HIDE_CHOICESboolFalseHides LLM choices (completions API outputs)

​
Embedding Controls

VariableTypeDefaultDescription
OPENINFERENCE_HIDE_EMBEDDINGS_VECTORSboolFalseReplaces embedding.embeddings.*.embedding.vector values with "__REDACTED__"
OPENINFERENCE_HIDE_EMBEDDINGS_TEXTboolFalseReplaces embedding.embeddings.*.embedding.text values with "__REDACTED__"
OPENINFERENCE_HIDE_EMBEDDING_VECTORSboolFalseDeprecated: Use OPENINFERENCE_HIDE_EMBEDDINGS_VECTORS instead

​
Size Limits

VariableTypeDefaultDescription
OPENINFERENCE_BASE64_IMAGE_MAX_LENGTHint32000Limits characters of a base64 encoding of an image. Images exceeding this length will be replaced with "__REDACTED__"

​
Usage Examples

​
Bash/Shell

# Hide all inputs and outputs
export OPENINFERENCE_HIDE_INPUTS=true
export OPENINFERENCE_HIDE_OUTPUTS=true

# Run your application
python app.py

​
Docker

FROM python:3.11

# Set environment variables
ENV OPENINFERENCE_HIDE_INPUT_TEXT=true
ENV OPENINFERENCE_HIDE_OUTPUT_TEXT=true
ENV OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH=16000

COPY . /app
WORKDIR /app
RUN pip install -r requirements.txt

CMD ["python", "app.py"]

​
Docker Compose

version: '3.8'
services:
  app:
    build: .
    environment:
      - OPENINFERENCE_HIDE_INPUTS=false
      - OPENINFERENCE_HIDE_OUTPUTS=false
      - OPENINFERENCE_HIDE_INPUT_IMAGES=true
      - OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH=16000

​
Kubernetes

apiVersion: v1
kind: ConfigMap
metadata:
  name: openinference-config
data:
  OPENINFERENCE_HIDE_INPUT_TEXT: "true"
  OPENINFERENCE_HIDE_OUTPUT_TEXT: "true"
  OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH: "16000"
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: my-app
spec:
  template:
    spec:
      containers:
      - name: app
        image: my-app:latest
        envFrom:
        - configMapRef:
            name: openinference-config

​
Python (.env file)

# .env
OPENINFERENCE_HIDE_INPUTS=false
OPENINFERENCE_HIDE_OUTPUTS=false
OPENINFERENCE_HIDE_INPUT_IMAGES=true
OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH=16000

# app.py
from dotenv import load_dotenv
load_dotenv()

# Environment variables are automatically picked up
from openinference.instrumentation.openai import OpenAIInstrumentor
OpenAIInstrumentor().instrument(tracer_provider=tracer_provider)

​
Node.js (.env file)

# .env
OPENINFERENCE_HIDE_INPUTS=false
OPENINFERENCE_HIDE_OUTPUTS=false
OPENINFERENCE_HIDE_INPUT_IMAGES=true
OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH=16000

// app.ts
import { config } from "dotenv";
config();

// Environment variables are automatically picked up
import { OpenAIInstrumentation } from "@arizeai/openinference-instrumentation-openai";
const instrumentation = new OpenAIInstrumentation();

​
Boolean Values

Boolean environment variables accept the following values (case-insensitive):
  • true, True, TRUE → true
  • false, False, FALSE → false
  • Any other value → defaults to false

​
Integer Values

Integer environment variables (like OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH):
  • Must be valid integer strings (e.g., "16000", "32000")
  • Invalid values will fall back to the default value
  • In Java, can also be set to "unlimited" for no limit

​
Common Scenarios

​
Development Environment

# Maximum observability - see everything
export OPENINFERENCE_HIDE_INPUTS=false
export OPENINFERENCE_HIDE_OUTPUTS=false

​
Production Environment

# Hide sensitive text but keep structure
export OPENINFERENCE_HIDE_INPUT_TEXT=true
export OPENINFERENCE_HIDE_OUTPUT_TEXT=true
export OPENINFERENCE_HIDE_EMBEDDINGS_TEXT=true

​
Compliance Requirements

# Hide all potentially sensitive data
export OPENINFERENCE_HIDE_INPUTS=true
export OPENINFERENCE_HIDE_OUTPUTS=true
export OPENINFERENCE_HIDE_EMBEDDINGS_VECTORS=true
export OPENINFERENCE_HIDE_EMBEDDINGS_TEXT=true

​
Reduce Payload Size

# Limit image sizes to reduce storage costs
export OPENINFERENCE_BASE64_IMAGE_MAX_LENGTH=8000

​
Redacted Content

When content is hidden due to environment variable settings, the value "__REDACTED__" is used as a placeholder. This allows trace consumers to identify that content was intentionally hidden rather than missing or empty.

​
Next Steps