arize-phoenix-evals

What it is and what it does

Phoenix Evals is a framework for building and running evaluations on language model applications. It provides both pre-built evaluators (for tasks like hallucination detection, relevance scoring, and tool invocation checking) and tools to compose custom evaluators using your choice of LLM provider. The package handles input mapping for complex nested data structures and integrates with OpenTelemetry for tracing and observability.

The framework is designed to work with pandas DataFrames for batch evaluation and supports both synchronous and asynchronous evaluation modes. It has nine runtime dependencies including jsonpath-ng, openinference-instrumentation, openinference-semantic-conventions, opentelemetry-api, pandas, pydantic, pystache, tqdm, and typing-extensions. The package is actively maintained, supports Python 3.10 through 3.14, and has no known security vulnerabilities.

Use it for:

• Detect hallucinations in LLM outputs by checking whether responses are grounded in provided context.
• Score retrieved documents for relevance to user queries in retrieval-augmented generation systems.
• Evaluate whether an LLM selected and invoked the correct tool with appropriate arguments.
• Run batch evaluations on large datasets of LLM interactions stored in pandas DataFrames.
• Build custom evaluators with templated prompts and multi-choice scoring for domain-specific assessment tasks.

Worth the install?

AI-flagged interpretation of the facts on this page — verify before relying

Phoenix Evals provides composable building blocks for evaluating LLM applications, including pre-built evaluators for hallucination detection, relevance, toxicity, and other common assessment tasks.

Yes, with conditions. The package is actively maintained, has low installation friction, and provides a practical framework for LLM evaluation with both pre-built and custom evaluators. However, the Elastic-2.0 license treatment is flagged as unclear—verify the license terms for your use case before production deployment. If you need LLM evaluation capabilities and can clarify the license, this is a solid choice.

Install

arize-phoenix-evals on PyPI

pip install arize-phoenix-evals

uv add arize-phoenix-evals

poetry add arize-phoenix-evals

Installing arize-phoenix-evals

Before you install

Low friction installation with a pure Python wheel. The package is actively maintained with a recent release and no known vulnerabilities, though the license treatment is unclear and may warrant review before production use.

License in practice

The package uses the Elastic-2.0 license, which is marked as having unclear treatment in the metadata. Review the license terms directly before deploying in commercial or proprietary contexts.

Quickstart

pip install arize-phoenix-evals

from phoenix.evals import create_classifier
from phoenix.evals.llm import LLM

llm = LLM(provider="openai", model="gpt-4o")
evaluator = create_classifier(
    name="helpfulness",
    prompt_template="Rate as helpful or not:\n\nQuery: {input}\nResponse: {output}",
    llm=llm,
    choices={"helpful": 1.0, "not_helpful": 0.0},
)
scores = evaluator.evaluate({"input": "How do I reset?", "output": "Go to settings > reset."})

Requires an LLM provider API key to run LLM-based evaluators; code-based evaluators work without external dependencies.

Verify before relying

• Whether Elastic-2.0 license permits commercial use without additional restrictions or obligations.
• Performance characteristics of the built-in concurrency and batching mentioned in the description.

Package facts

Evidence: arize_phoenix_evals-3.4.0-py3-none-any.whl

Keywords: Explainability, Monitoring, Observability

Tags

More Artificial Intelligence packages

Further reading