Arize: LLM Observability and Evaluation Platform

Key Features

• Agent Tracing

  Visualizes the flow of data through AI applications to identify bottlenecks and understand agent paths.

• LLM Evaluations

  Supports online and offline assessments of task performance using LLM-as-a-Judge and custom templates.

• Prompt Playground

  Allows users to test prompt changes and view feedback against different datasets.

• Drift Detection

  Monitors feature and embedding drift across training and production environments to identify performance shifts.

• Human Annotation

  Provides workflows for identifying and correcting errors and managing labeling queues for dataset creation.

• Production Monitoring

  Uses dashboards and monitors to surface issues such as hallucinations or PII leaks.

Use Cases

• AI Agent Debugging

  Tracing multi-agent interactions to identify where a process failed or where a hallucination occurred.

• Prompt Iteration

  Using the prompt playground to compare different prompt versions against a golden dataset.

• Production Guardrails

  Setting up monitors to detect PII leaks or performance regressions in a live AI application.

• Model Performance Analysis

  Using heatmaps and cluster search to identify underperforming data slices in ML models.

Integrations

• OpenTelemetry

FAQ

What is Arize used for?

Arize is used to monitor, debug, and evaluate AI agents and LLM applications. It helps teams trace data flows and detect performance issues like hallucinations or model drift.

Does Arize have a free version?

Yes, Arize offers a free open-source version called Phoenix and a free SaaS tier (AX Free) for individuals and startups.

Who is the target user for Arize?

The platform is designed for AI engineers, data scientists, and software companies building production-ready AI agents.