Arize Phoenix Features

Enables step-by-step inspection of agent reasoning, tool calls, and intermediate outputs to identify failure points.

Displays agent execution paths, decision trees, and tool call sequences in a visual, navigable format.

Traces multi-step AI agent workflows including LLM calls, tool invocations, retrieval steps, and intermediate reasoning states.

Detects behavioral regressions by comparing agent performance across model versions, prompt changes, or configuration updates.

Identifies instances where the agent generates factually incorrect or unsupported content.

Assesses agent output quality using methods such as LLM-as-judge, rule-based checks, or custom evaluators.

Notifies teams of quality degradations, error spikes, or anomalous agent behavior in production.

Tracks response times across agent steps, LLM calls, and tool invocations to identify performance bottlenecks.

Monitors token consumption and attributes costs across models, agents, and workflows.

Maintains comprehensive logs of agent actions, decisions, and data access for compliance and governance purposes.

Automatically identifies and inventories AI agents deployed across environments, including their configurations and dependencies.

Monitors agent behavior against defined organizational policies, guardrails, and regulatory requirements.