When to use Arize AX or Phoenix
Choose Phoenix for open-source control, Arize AX for managed production workflows, or use both across development and production.
Phoenix and Arize AX at a glance
Arize Phoenix
Free. Open source. Self-hosted.
Run AI observability and evaluation in infrastructure you control.
Run it in your stack
Local, Docker, Kubernetes, private cloud, or air-gapped.
Work close to the code
Trace, evaluate, test prompts, and run experiments.
Keep full control
You choose where it runs, how it’s configured, and when it changes.
Operate it your way
Your team manages scaling, upgrades, security, and availability.
Best for teams that want open-source control and have the engineering capacity to operate the platform.
Arize AX
Commercial. Managed SaaS or self-hosted.
Run shared production observability, evaluation, and improvement workflows in a commercially supported platform.
Choose your deployment
Managed SaaS or self-hosted.
See production behavior
Bring traces, evaluations, alerts, and experiments together.
Run the loop continuously
Find recurring failures, investigate issues, and verify changes.
Coordinate across teams
Centralize governance, access, evidence, and support.
Best for teams that need continuous production workflows, shared governance, or managed operations.
From trace to tested fix
Phoenix and AX support the same AI engineering loop: inspect, evaluate, test, ship, and repeat.
Phoenix is open source and operated by your team. AX is a supported platform for shared production workflows.
- 01 Inspect the trace
- 02 Evaluate behavior
- 03 Build the dataset
- 04 Test the change
- 05 Ship and measure
Production results feed the next iteration.
Which one should you use?
Arize Phoenix
Best when
- You want an open-source, self-managed platform
- Your team wants direct access to the platform and APIs
- Your platform or SRE team can operate the service
You gain
Control over your data, deployment, integrations, and release cadence.
Your team owns
Infrastructure, PostgreSQL, scaling, upgrades, security, uptime, and production automation.
Phoenix can remain your permanent production platform.
Arize AX
Best when
- Production quality workflows must run continuously
- Multiple teams need shared workflows and governance
- You need managed compute, commercial support, or SLAs
You gain
A centralized production system without assembling every workflow yourself.
What stays yours
Instrumentation quality, data policies, evaluation criteria, and the definition of good application behavior.
Start with AX when these requirements already exist on day one.
Which setup fits your workflow?
Start with where your team needs speed, control, or shared production operations. Phoenix and AX each cover two of these situations, and a hybrid of the two covers development and production at once.
Debug a failing agent now
Use Phoenix locally
Inspect traces, turn failures into datasets, and test fixes in a tight local loop.
Own the stack end to end
Use self-hosted Phoenix
Keep telemetry in your environment and control deployment, APIs, upgrades, and production automation.
Coordinate quality across teams
Use Arize AX
Centralize production telemetry, evaluations, monitoring, and experiments so teams can find failures and verify fixes together.
Keep AX inside your boundary
Use self-hosted Arize AX
Run AX in your own infrastructure, with your team owning the environment while Arize licenses and supports the platform.
Developers can also keep Phoenix locally while the platform team runs production in AX.
Compare capabilities at a glance
A quick view of what each product supports, and where the operating model differs.
- Available
- Available, team-operated or self-assembled
- Not a primary product capability
| Capability | Phoenix | Arize AX | Notes |
|---|---|---|---|
| Core workflow | |||
| Tracing | Same OpenTelemetry and OpenInference span semantics | ||
| Evaluations | Phoenix provides the primitives; AX adds managed runs on production data | ||
| Datasets | Collect representative examples in either product | ||
| Experiments | Managed as part of the production loop in AX | ||
| Prompt management | Version and iterate inside the platform | ||
| Replay workflows | Included in a self-hosted Phoenix instance | ||
| Observability and production operations | |||
| Dashboards | Team-built in Phoenix; custom dashboards in AX | ||
| Metrics | Team-built via Phoenix APIs; managed in AX | ||
| Monitors | Team-built via Phoenix APIs; managed in AX | ||
| Alerts | Team-built via Phoenix APIs; managed in AX | ||
| Labeling workflows | Available across current AX plans | ||
| Production issue discovery | Signal in AX; team-built scheduled reviews in Phoenix | ||
| Agentic and AI-assisted workflows | |||
| Interactive AI engineering agent | PXI inside Phoenix; Alyx inside AX | ||
| Agentic evaluation | Sandboxed evaluations in Phoenix; Agent-as-a-judge managed in AX | ||
| Managed agent workflows | MCP, skills, and CLI in Phoenix; Managed Agents hosted in AX | ||
| Full-agent experiments | ATIF in Phoenix; managed inside the AX production loop | ||
| Platform and development | |||
| Local development | Phoenix runs on a laptop or in Docker | ||
| Self-hosted deployment | Free in Phoenix; commercial license for AX | ||
| Managed SaaS deployment | Arize operates the infrastructure | ||
| Air-gapped deployment | Phoenix runs on tightly restricted networks | ||
| OpenTelemetry and OpenInference | Same instrumentation routes to either backend | ||
| API access | Direct platform APIs in both products | ||
| Access controls | Instance-level in Phoenix; organization-level in AX | ||
| Retention policies | Team-set in Phoenix; configured in AX | ||
| Governance for multiple teams | Organization, workspaces, and auditing in AX | ||
| Commercial support and SLAs | Community support and internal expertise in Phoenix | ||
Online evaluations, dashboards, metrics, monitors, labeling workflows, and experiments are available across current AX plans.
More advanced agent, security, and deployment capabilities depend on plan and deployment.
Everything you need to know about Arize AX and Phoenix.
Start with your first trace
Arize Phoenix
Run Phoenix locally
Launch Phoenix in your environment and begin debugging traces close to the code.
uvx arize-phoenix serve
Arize AX
Start free with Arize AX
Create an AX account, instrument your agent, and send your first trace from the terminal.
npx evals
Using Phoenix and AX together?
Start with Phoenix close to the code, then route production telemetry to AX.