Phoenix by Arize

Phoenix by Arize is a free, open-source AI observability and evaluation platform for engineering teams that need OpenTelemetry-aligned tracing, LLM and agent debugging, prompt experiments, datasets, evaluator workflows, and a managed upgrade path through Phoenix Cloud or Arize AX when self-hosted operations are no longer enough. The core Phoenix project is designed for teams building production AI systems where normal application logs are insufficient: it captures span-level detail across LLM calls, retrieval steps, tool invocations, prompt templates, variables, model responses, evaluator scores, token usage, and custom application logic.

Phoenix is strongest when a team wants to understand why an LLM or agent workflow produced a specific result, then turn that evidence into repeatable evaluation and improvement loops. Developers can instrument applications with Python or JavaScript SDKs, OpenInference, or OpenTelemetry-compatible spans, then inspect traces in Phoenix to see the full execution path. That makes it useful for debugging multi-step agents, reviewing retrieval-augmented generation behavior, comparing prompt variants, building datasets from real traces, and scoring outputs with LLM-as-judge, code-based checks, or human labels. Because Phoenix is aligned with OpenTelemetry OTLP rather than a closed tracing format, it fits teams that care about portability and interoperability across observability stacks.

Several concrete facts matter for buyers comparing Phoenix with managed AI observability products. Phoenix self-hosted is free and open source, but the team running it owns infrastructure, retention, upgrades, scaling, and storage costs. Phoenix Cloud provides 2 free hosted Phoenix instances, each preconfigured with 10 GiB of storage. There is no published paid Phoenix Cloud plan, paid instance price, or paid storage overage schedule; teams that need more managed production capacity are directed toward Arize AX or enterprise discussions rather than a metered Phoenix Cloud upgrade. Arize AX Free includes 25k trace spans per month, 1 GB ingestion volume per month, and 15 days retention. Arize AX Pro is listed at $50 per month and includes 50k trace spans per month, 10 GB ingestion volume per month, 30 days retention, higher rate limits, and email support. Arize also reported in June 2026 that Phoenix reached 10,000 GitHub stars, and its 2026 site describes millions of monthly downloads, indicating meaningful open-source adoption.

The practical tradeoff is control versus managed convenience. Phoenix OSS is a strong starting point for engineering-led teams that want local development, Docker, Kubernetes, or self-hosted cloud deployment without committing to a SaaS bill. Arize AX is the clearer fit when the organization needs hosted infrastructure, online evaluations, product observability monitors, custom metrics, longer retention, email or enterprise support, the Alyx AI debugging assistant, and contracted security or compliance controls. Phoenix is not a no-code analytics product, and its evaluation quality depends on the team's datasets, labels, scoring criteria, and review process. For teams willing to instrument their systems and define what good output means, it provides a deep, standards-aligned workflow for tracing, evaluating, debugging, and improving LLM applications and AI agents.