Tracing and trace setup
Capture application traces to see what happened inside each request. The docs describe manual or auto-instrumentation and tracing setup with 30+ providers and frameworks.
Arize AX
RivendicaArize AX is an AI engineering platform for tracing, evaluating, and improving AI agents and AI apps with observability, experimentation, and AI-assisted debugging.
AI engineering platform for agent observability
Arize AX is an AI engineering platform for observing, evaluating, and improving AI agents and AI applications. The homepage describes it as a continual learning platform that helps teams trace what their systems did, measure quality with evaluations, and iterate on prompts, models, and workflows before and after production.
The product sits alongside Phoenix, Arize’s open-source observability and evaluation tool. AX builds on open standards such as OpenInference and OpenTelemetry, and the site positions it for AI engineers and product teams that need tracing, experimentation, dashboards, and AI-assisted debugging in one platform.
Core capabilities
Trace investigation
Find problematic traces using queries, filters, and AI Search, then inspect span, trace, and session data for debugging and investigation.
Evaluation workflows
Create evaluators, run offline and online evaluations, and use human review, labeling queues, and dashboards to monitor quality over time.
Experimentation and prompt iteration
Curate datasets, store experiment runs, compare prompts and models, and gate deployment with CI/CD based on experiment performance.
AI-assisted engineering with Alyx
Use Alyx to debug traces, run evals, compare experiments, suggest prompt edits, and generate dashboards from natural-language requests.
Open standards and integrations
Support OpenInference and OpenTelemetry-based instrumentation across a broad integration surface, including LLM providers, agent frameworks, coding agents, and orchestration tools.
Common use cases
Debugging live agent behavior
Instrument an AI application, capture traces, and inspect spans to understand what happened inside each request when debugging failures or slow paths.
Monitoring model and agent quality
Run evaluators against traces, spans, or sessions to measure quality continuously and track whether changes are improving or degrading performance.
Pre-release experimentation
Compare prompts, models, or experiment runs on curated datasets before releasing changes to production and use CI/CD gates where needed.
Human-in-the-loop evaluation
Use human annotation, feedback tracking, and aligned evaluators to calibrate automated scores against real judgments and identify failure patterns.
Integrating with an existing AI stack
Adopt open-standard tracing across frameworks and providers so teams can connect their existing stack without switching to a proprietary format.
Pros and Cons
Pros
- Covers tracing, evaluations, experimentation, and prompt iteration in one product.
- Supports both managed SaaS and self-hosted enterprise deployment options.
- Built on open standards, including OpenInference and OpenTelemetry.
- Supports a broad integration surface across models, frameworks, tools, and environments.
- Includes Alyx, an AI engineering agent for debugging, evals, and workflow assistance.
Cons
- The source does not provide full documentation for every feature, so some implementation details remain unclear from the public pages alone.
- Pricing and deployment specifics vary by plan, and some advanced capabilities are only described at a high level on the site.
FAQ
How do I get started with Arize?
You connect an agent or application to Arize and send your first trace. The platform uses traces to show what happened inside each request, and the docs note that you can start setting up traces and then use Alyx for help.
Does Arize work with my existing stack?
Yes. The docs say Arize integrates with 40+ models, frameworks, and AI tools, including OpenAI, Anthropic, Google, Amazon Bedrock, LangGraph, LangChain, LlamaIndex, CrewAI, OpenAI Agents SDK, and DSPy.
Can Arize be deployed outside a single cloud?
Arize supports flexible deployment options and the pricing page lists SaaS and self-hosted options for enterprise. The FAQ and pricing page also point to deployment across Google Cloud, AWS, Azure, and self-hosted environments.
Is Arize meant for generative AI applications?
Yes. The site says Arize is built for modern AI applications including chatbots, RAG systems, copilots, and agents, with workflows for tracing, evaluation, monitoring, and iteration.
What is the difference between Arize AX and Phoenix?
Phoenix is the open-source product for tracing, evaluation, experimentation, and prompt iteration, and the site says it can run locally or self-hosted. Arize AX is the managed enterprise platform on the same open standards, with managed infrastructure and additional workflows.
Quick Facts
- Category
- AI engineering platform
- Primary users
- AI engineers and product teams
- Deployment
- SaaS and self-hosted options
- Standards
- OpenInference and OpenTelemetry
- Source domain
- arize.com
- Pricing model
- Free, Pro, and Enterprise plans are listed