%3Aquality(90)%2F&w=3840&q=75)
Capabilities | Arize AX | |
|---|---|---|
What you investigate | Agent conversations linked to traces, alongside service metrics and logs. Reduce mean time to resolution by connecting AI failures to the services and dependencies that caused them. Learn more | Arize AX AX traces model, retrieval, and tool operations; it also provides production monitors. |
OpenTelemetry support | The Agent Observability SDK emits OpenTelemetry traces and metrics, plus a separate generation-data export. You can also send compatible generative AI spans without the SDK; OpenTelemetry ingestion is in public preview. Learn more | Arize AX Uses OpenTelemetry and OpenInference attributes, exported through OTLP. |
Optimize cost and performance | Agent Observability estimates the cost of every generation from its token usage and a price for the model that served it. By default, it uses public list prices from the model card catalog. If you pay a different negotiated price, set your own price for that model. Learn more | Arize AX You can add new cost tracking or edit cloned defaults to apply negotiated discounts or create tiered pricing cost configs for volume-based tiered pricing. |
Evaluation & experiments | Score live traffic with online evaluations, including built-in safety and secret-detection templates, and alert on score drops. Compare agent versions before release with offline experiments you run yourself, including from CI. | Arize AX Online evaluations on traces and offline dataset/experiment comparisons, including runs wired into CI/CD. |
Prompt versioning | Tracks agent versions and prompt changes reported by your instrumentation (or derived from the system prompt) and compares quality, cost, and latency across versions. | Arize AX Prompt Hub saves each prompt change as an immutable, versioned Prompt Object, and versions can be diffed. |
Dashboards & alerts | Built-in dashboards and Grafana alerts for agent metrics. Learn more | Arize AX Custom metrics and monitors with static or automatic thresholds and notifications. |
Usage and billing | Bills model calls as generations, which is one model call made by your agent, plus evaluation/guard tokens and telemetry. Learn more | Arize AX AX plans set span-volume, ingestion, retention, and Signal limits. Model-cost tracking is a separate view of your application spend. |
Ecosystem & integrations | SDKs for Go, Python, TypeScript, Java, and .NET; integrations include LangChain, LangGraph, OpenAI Agents, and Vercel AI SDK. Learn more | Arize AX Automatic integrations and manual OTel instrumentation; OpenInference supplies AI-specific attributes. |
AI-assisted investigation | Assistant can inspect calls and help configure evaluations. Alert Enrichment can start an investigation, call an external service, or query a Grafana data source when an alert fires. Learn more | Arize AX Alyx helps investigate and edit. Signal scans traces for recurring issues; repository-backed fixes and pull requests require Enterprise. |
Hosting | Agent Observability runs in Grafana Cloud. Setup requires enabling it on a Cloud stack. Learn more | Arize AX AX offers managed SaaS and Enterprise self-hosting; Phoenix can also be operated in production. |
Setup | Install the gcx plugin, then use its agento11y-instrument skill to instrument your app and send generations to Agent Observability. The skill verifies that data lands in Grafana Cloud. Learn moreOpens in new tab | Arize AX Send OTel spans with OpenInference attributes for AI-specific views. |

%3Aquality(90)%2F&w=3840&q=75)
%3Aquality(90)%2F&w=3840&q=75)
%3Aquality(90)%2F&w=3840&q=75)
%3Aquality(90)%2F&w=3840&q=75)