Arize Phoenix is an open-source AI observability platform for LLM applications. It provides tracing, evaluation, datasets, and experiments to help you build and improve AI applications.

Services

phoenix: The main Phoenix application server with UI and OpenTelemetry collectors.
phoenix-db: PostgreSQL database for persistent storage.

Ports

Port	Protocol	Description
6006	HTTP	UI and OTLP HTTP collector (`/v1/traces`)
4317	gRPC	OTLP gRPC collector

Environment Variables

Variable Name	Description	Default Value
PHOENIX_VERSION	Phoenix image version	`version-12.19.0`
PHOENIX_PORT_OVERRIDE	Host port for Phoenix UI and HTTP API	`6006`
PHOENIX_GRPC_PORT_OVERRIDE	Host port for OTLP gRPC collector	`4317`
PHOENIX_ENABLE_PROMETHEUS	Enable Prometheus metrics endpoint	`false`
PHOENIX_SECRET	Secret for authentication (optional)	`""`
POSTGRES_VERSION	PostgreSQL image version	`17.2-alpine3.21`
POSTGRES_USER	PostgreSQL username	`postgres`
POSTGRES_PASSWORD	PostgreSQL password	`postgres`
POSTGRES_DB	PostgreSQL database name	`phoenix`

Volumes

phoenix_db_data: PostgreSQL data volume for persistent storage.

Getting Started

Copy the example environment file:
```
cp .env.example .env
```

(Optional) For production, set a secure password and secret:

# Generate a secret for authentication
openssl rand -base64 32

Start the services:
```
docker compose up -d
```
Access Phoenix UI at http://localhost:6006

Sending Traces

Phoenix supports OpenTelemetry-compatible traces. You can send traces using:

HTTP (OTLP)

Send traces to http://localhost:6006/v1/traces

gRPC (OTLP)

Send traces to localhost:4317

Python Example

from phoenix.otel import register

tracer_provider = register(
    project_name="my-llm-app",
    endpoint="http://localhost:6006/v1/traces",
)

Features

Tracing: Capture and visualize LLM application traces with OpenTelemetry support.
Evaluation: Run evaluations using built-in or custom evaluators.
Datasets: Create and manage datasets for testing and evaluation.
Experiments: Run experiments to compare model performance.
Playground: Test prompts with different models interactively.

Documentation

For more information, visit the official Phoenix documentation.

Security Notes

Change default PostgreSQL password in production.
Set PHOENIX_SECRET for authentication if exposing Phoenix publicly.
Consider using a reverse proxy with SSL/TLS in production.
Regularly backup the PostgreSQL database.