Arize AI

AI Observability and Evaluation

Arize is an AI observability and evaluation platform. It helps engineers build, evaluate, and monitor AI applications and agents. Development teams use Arize to run experiments in development, and trace and evaluate performance in production.

Arize has built several OSS packages to support this goal:

• Phoenix — an open source AI observability platform for developers

• OpenInference — an open source instrumentation package to trace LLM applications across models and frameworks.

We log over 1 trillion inferences and spans, 10 million evaluation runs, and 2 million OSS downloads every month.

Features

💬 Prompts

Iterate on prompts

• Prompt playground - compare different prompts side by side

• Prompt hub - manage and version your prompts in one place

• Prompt builder - generate prompts with AI

• Save as experiment - systematically A/B test prompts against large datasets

🧪 Experiments

Run experiments

• Curate datasets - create and update test datasets to measure performance
• Track experiments - store every experiment run in a structured format
• Evaluate experiments - systematically measure performance improvements based on LLM and code evaluations
• CI/CD - gate deployment to production based on experiment performance

🔀 Tracing

Trace your application

• Setup tracing instrumentation - get instant visibility into your application traces
• Find problematic traces - use our search and filter capabilities to find outliers of poor performance
• Run quick evaluations - determine the causes of poor performance across hundreds of spans
• Powered by Phoenix OSS

✅ Evaluation

Evaluate performance

• Evaluate production data - run evals continuously against your data
• Track key metrics - create custom dashboards to monitor performance
• Get alerts - get alerts when performance deviates from the norm
• Guardrail bad outputs - prevent poor performing outputs from reaching users
• Annotate your outputs - use labeling queues to run evals and annotate your spans in one place

✨AI Copilot

Build AI with AI

• AI Search - find patterns in your data
• Create Custom Evaluations - write tailored evals based on custom criteria
• Diagnose RAG Issues - analyze your document retrieval and suggest improvements
• Span Chat - analyze and evaluate any span in chat
• Dashboard Generator - generate dashboard widgets with natural language
• Optimize Prompts - get suggested prompt edits based on best practices

And more!

Quickstarts

Running Arize for the first time? Select a quickstart below.

Run experiments

Systematically experiment and evaluate performance against your data.

Trace your application

Get visibility into your application traces using our auto-instrumentation plugins.

Iterate on prompts

Iterate on prompt templates, input variables, LLM models, and parameters without code.

Run evaluations

Run LLM or code evaluations continuously against your data.

Next Steps

Try our tutorials

Check out a comprehensive list of example notebooks for agents, RAG, voice, tracing, evals, and more.

Watch our paper readings

See our video deep dives on the latest papers in AI.

Join our Slack community

Join the Arize Slack community to ask questions, share findings, provide feedback, and connect with other developers.

Machine Learning Guides

Looking for traditional machine learning or CV guides? Start here:

Machine Learning

Log inferences and debug your machine learning models

Computer Vision

Run similarity search and evaluate performance of your CV models