Arize

Observe, debug, and improve AI and LLM systems in production.

Arize is an AI observability and evaluation platform that helps teams monitor the performance of machine learning models and LLM-powered applications after they are deployed. It surfaces issues such as model drift, data quality problems, prompt regressions, and hallucinations by tracking inputs, outputs, and performance metrics over time. Arize also provides evaluation tooling that lets teams run systematic tests against their LLM applications to measure quality before and after changes. The platform serves a range of company sizes, from startups to enterprises, and includes Phoenix, an open-source observability library for teams that want a self-managed option. It differentiates itself through its combined focus on both traditional ML monitoring and the newer challenges specific to LLM and agent systems.

Compliance

GDPR HIPAA SOC 2

Visit Arize

Key Features

LLM tracing and monitoring: Captures detailed traces of LLM application calls — including prompts, completions, latency, and token usage — to give teams visibility into how their applications are behaving in production.
Evaluation framework: Provides tools to run automated evaluations of LLM outputs, helping teams measure quality metrics like relevance, groundedness, and toxicity systematically rather than ad hoc.
Drift and degradation detection: Monitors model and application performance over time and alerts teams when outputs degrade, data distributions shift, or error rates increase.
Agent observability: Tracks multi-step agent workflows, capturing the sequence of tool calls, model invocations, and decisions so teams can debug complex agent behavior.
Phoenix open-source library: Offers an open-source version of its observability tooling that developers can run locally or self-host, lowering the barrier to adoption.

Use Cases

For ML and AI teams at product companies: An AI product team at a company like a marketplace or delivery platform uses Arize to monitor their recommendation or search models alongside newer LLM features, keeping all model observability in one platform.
For teams deploying LLM-powered applications: A team that has shipped an AI assistant to customers uses Arize to track whether answer quality is holding up over time, catch prompt regressions after model updates, and evaluate improvements before releasing them.
For developers building agentic systems: An engineering team building a multi-step research agent uses Arize to trace what each agent step is doing, identify where failures occur, and understand the cost and latency profile of their pipeline.

Pricing MODELS

Freemium

Pricing Summary

Arize offers a free tier for getting started, a Team plan at $100/month, and custom enterprise pricing for larger organizations or those requiring on-premise deployment. The Phoenix open-source library is free to use and self-host. Full pricing is available on their website.