Arize AX (Arize AI, 2020) — managed AI engineering platform: Observe + Evaluate + Develop + Alyx + Signal + adb. Framework-agnostic, built on OpenInference + OpenTelemetry, 40+ integrations (OpenAI, Anthropic, LangGraph, LangChain, LlamaIndex, CrewAI). SaaS US/EU/CA or self-hosted. 1T spans/mo scale, customers Reddit, Uber, Spotify.

Arize AX is Arize AI's commercial managed AI engineering platform, positioned as a continual learning platform for agents — a comprehensive stack covering the entire AI application lifecycle: observe, evaluate, develop, and autonomously improve (Alyx / Signal). Built on OpenInference (Arize's own open standard of semantic conventions) and OpenTelemetry — instrument once, no proprietary trace format. Framework-agnostic: works with OpenAI, Anthropic, Google, Amazon Bedrock, LangGraph, LangChain, LlamaIndex, CrewAI, OpenAI Agents SDK, DSPy, and 40+ other stacks. An open-source version (Phoenix) is available under Apache 2.0.
Four product pillars of Arize AX. (1) Observe — OpenTelemetry-compliant tracing of every LLM/tool/agent call, token/latency/cost tracking, session support, multi-modal tracing (image, voice, PDF), agent trajectory visualisation (path and graph), custom dashboards/metrics/monitors/views, datalake sync with adb DataFabric. (2) Evaluate — online and offline evals on traces/spans/datasets/experiments (unlimited across all plans), custom code evaluators (Enterprise), Agent-as-a-Judge, Align evaluators, Trace Evals (agent trajectories), Session Evals (multi-turn conversations), multi-modal evaluation, human annotations, and labeling queues.
(3) Develop — unlimited Playgrounds, prompt versioning and serving, trace replay in playground, multi-prompt comparison, agent experimentation, Prompt Learning (PL) optimisation, unlimited datasets and experiments. (4) Improve (Alyx + Signal + Agent Swarms) — Alyx is an in-app AI engineering agent (like Cursor/Claude Code for AI engineering) running evals, debugging issues, and improving customer agents; Signal is always-on issue detection with auto-generated PRs; Agent Swarms orchestrate sandboxed managed debugging agents and provide swarm observability across managed and third-party agents. Enterprise adds Agent-as-a-Judge and dedicated engineers.
adb (Arize datastore) is a purpose-built datastore for GenAI traces — per Arize's own benchmarks, the most scaled platform for storing agent trajectories and context. Open formats, native integration with BigQuery, Databricks, Snowflake via DataFabric. Arize's operational scale: 1 trillion spans/month, 1 billion evals/month, 5 million Phoenix downloads/month. adb and DataFabric available from Enterprise tier.
Three plans + Startup program. AX Free ($0, 25k spans/mo, 1 GB storage, 15-day retention, unlimited users and evals, community support). AX Pro ($50/month, 50k spans/mo, 10 GB storage, 30-day retention, email support). AX Enterprise (custom pricing, custom spans/storage/retention, SaaS or self-hosted, RBAC, SSO, audit logs, HIPAA, SLA, Alyx, Signal, Agent Swarms, adb DataFabric, dedicated support). Startup pricing available separately.
Enterprise-grade security. Certifications: SOC 2 Type II, ISO 27001, PCI DSS, HIPAA, GDPR (Trust Center: arize.com/trust-center). Data residency: US, EU, Canada for SaaS; self-hosted on AWS/GCP/Azure Kubernetes in the customer's own VPC (Enterprise). Customers: Reddit, DoorDash, Instacart, Uber, Spotify, Booking.com, TripAdvisor, PagerDuty, Snorkel, Handshake, Atlassian, PepsiCo, TheFork, Wayfair, Typeform, Priceline, Air Canada, Cohere, Conde Nast, BenchSci, BazaarVoice, Atropos, Defense Innovation Unit. Competition: LangSmith (LangChain), Braintrust, DeepEval / Confident AI (open-source), Weights & Biases Weave, Fiddler AI, Langfuse. Differentiator: OpenInference as an open standard plus Phoenix (open-source, 5M+ downloads) build a broader foundation than competitors.