Arize AI

Arize AI builds the Arize AX platform — a unified environment for developing, evaluating, and observing AI agents and LLM applications in production. Founded by Jason Lopatecki and Aparna Dhinakaran, the company is backed by Battery Ventures, Foundation Capital, TCV, Microsoft Ventures, and others. The platform is available as a cloud SaaS offering and as a self-hosted option, with an open-source counterpart called Phoenix.

What It Is

Arize AX is an AI and agent engineering platform that closes the loop between development and production. It covers three core pillars: development tooling (prompt management, optimization, and playground), evaluation (LLM-as-a-Judge, CI/CD experiments, human annotation), and observability (OpenTelemetry-based tracing, online evals, monitoring dashboards). The platform also includes Alyx, an AI engineering agent that helps teams debug faster and build with greater confidence. A purpose-built datastore called adb underpins the platform, designed for real-time ingestion, sub-second queries, and elastic compute at petabyte scale.

Core Capabilities

• Open Standard Tracing: Agent and framework tracing powered by OpenTelemetry (OTEL) and the OpenInference conventions, making it vendor-, framework-, and language-agnostic.
• LLM as a Judge: Automated evaluation of prompts and agent actions at scale using LLM-based evaluators, all open-source.
• CI/CD Experiments: Regression detection for prompts and agents integrated into development pipelines.
• Human Annotation & Queues: Labeling queues, production annotations, and golden dataset creation in one place.
• Prompt Hub: Prompt serving, management, environment tags, multi-prompt comparison, and automatic optimization.
• ML & Computer Vision Observability: Model drift detection, embedding monitoring, heatmap-based failure analysis, and data curation for traditional ML and CV models.
• Alyx: A context-aware AI agent that assists with span/trace debugging, dashboard creation, prompt optimization, and AI trace search.

Open-Source Foundation

Arize publishes Phoenix, a fully open-source, self-hostable observability and evaluation tool for LLM applications, available on GitHub. The evaluation libraries and eval models are also open-source. The platform is built on OpenTelemetry and uses standard data file formats to avoid data lock-in. According to Arize, Phoenix reaches approximately 5 million downloads per month.

Platform Architecture and Deployment

Arize AX is available as a multi-tenant SaaS with data region options (US, EU, or CA) and as a self-hosted enterprise deployment supporting data residency and multi-region configurations. The underlying adb datastore is purpose-built for generative AI workloads, handling real-time ingestion and sub-second analytical queries. The enterprise tier adds SOC2 Type II, HIPAA compliance, SSO enforcement, space-level RBAC, audit logs, and uptime SLAs.

Scale and Adoption Signals

Arize's homepage states the platform has processed 1 trillion spans and runs 50 million evaluations per month. The vendor's customer page lists organizations including DoorDash, Instacart, Reddit, Uber, Booking.com, PagerDuty, Siemens, PepsiCo, TripAdvisor, and others as customers. The Defense Innovation Unit page notes that the U.S. Navy awarded prototype agreements to Arize AI as part of Project AMMO for ML-based maritime target recognition.

Why It Matters for AI Teams

The platform addresses a core challenge in production AI: without visibility into agent accuracy, hallucinations, and model drift, teams cannot reliably improve or trust their systems. By integrating tracing, evaluation, and prompt management in one place — and connecting production data back to the development loop — Arize AX enables what the company describes as a "data-driven iteration cycle." The inclusion of Alyx as an in-platform AI agent, and the open-source Phoenix project as a free self-hosted entry point, positions Arize across both individual developers and large enterprise engineering teams.