Breaking Change: CLI Now Subcommand-First#

April 7, 2026

Breaking change in arize-phoenix 14.0.0

The Phoenix server CLI is now subcommand-first. Flags that previously preceded the subcommand must now follow it.

Before:

phoenix --dev servephoenix --host 0.0.0.0 --port 6006 servepython -m phoenix.server.main --dev --dev-vite-port 5173 serve

After:

phoenix serve --devphoenix serve --host 0.0.0.0 --port 6006python -m phoenix.server.main serve --dev --dev-vite-port 5173

The db migrate subcommand is unchanged. Pass --database-url directly to the subcommand that needs it, or rely on PHOENIX_SQL_DATABASE_URL:

phoenix serve --database-url "postgresql://..."phoenix db migrate --database-url "postgresql://..."

The deprecated --enable-websockets flag has been removed. Use phoenix serve --help or phoenix db migrate --help for full option listings.

Breaking Change: Legacy px.Client() Removed#

April 7, 2026

Breaking change in arize-phoenix 14.0.0

phoenix.session.client.Client (accessed as px.Client()) has been removed. All client interactions now go through arize-phoenix-client.

pip install arize-phoenix-client

Before:

import phoenix as pxclient = px.Client(endpoint="http://localhost:6006")spans_df = client.get_spans_dataframe()

After:

from phoenix.client import Clientclient = Client(base_url="http://localhost:6006")spans_df = client.spans.get_spans_dataframe()

The endpoint parameter is now base_url. When omitted, the client falls back to environment variables or http://localhost:6006. The new client organizes methods under resource namespaces (.spans, .traces, .datasets, .experiments) instead of flat methods on the client object.

See the v14 migration guide for a full method-by-method mapping.

Breaking Change: /v1/evaluations Endpoint Removed#

April 7, 2026

Breaking change in arize-phoenix 14.0.0

The POST /v1/evaluations and GET /v1/evaluations REST endpoints have been removed. Use the annotations API instead:

Before:

from phoenix.trace import SpanEvaluationsimport phoenix as pxpx.Client().log_evaluations(    SpanEvaluations(eval_name="Hallucination", dataframe=results_df))

After:

from phoenix.client import ClientClient().spans.log_span_annotations_dataframe(    dataframe=results_df,    annotation_name="Hallucination",    annotator_kind="LLM",)

protobuf is no longer a direct dependency of the Phoenix server (it remains a transitive dependency via OpenTelemetry gRPC packages).

Breaking Change: Evals 1.0 Removed#

April 7, 2026

Breaking change in arize-phoenix-evals 3.0.0, arize-phoenix-client 2.3.1+

arize-phoenix-evals 3.0.0 removes the legacy evals 1.0 module and the legacy experiments module from the Phoenix server package.

Removed from arize-phoenix-evals:

• The entire legacy/ subpackage and its models/ wrappers
• MultimodalPrompt, PromptPartContentType, and PromptPart types — all adapter methods now use PromptLike

Removed from arize-phoenix:

• phoenix.experiments — the legacy experiment execution module (functions.py, tracing.py, evaluators/)
• phoenix.experiments.types — use phoenix.client.__generated__.v1.DatasetExample instead

Experiments migration:

# Beforefrom phoenix.experiments.types import Examplefrom phoenix.experiments.evaluators import create_evaluatorfrom phoenix.experiments import run_experiment, evaluate_experimentexperiment = run_experiment(dataset, task, evaluators=[...])# Afterfrom phoenix.client.__generated__.v1 import DatasetExample as Examplefrom phoenix.client.experiments import create_evaluator, run_experiment, evaluate_experimentexperiment = run_experiment(dataset=dataset, task=task, evaluators=[...])

The phoenix-client experiments module (phoenix.client.experiments) is the replacement and has no dependency on arize-phoenix-evals.

Breaking Change: GraphQL Forward Pagination Requires first#

April 7, 2026

Breaking change in arize-phoenix 14.0.0

Three GraphQL connection fields now require an explicit first argument and no longer accept backward pagination (last/before):

Before:

query {  project(id: "...") {    spans {      edges { node { name } }    }  }}

After:

query {  project(id: "...") {    spans(first: 100) {      edges { node { name } }    }  }}

Queries that omit first will fail with "first" is required. Queries that pass first greater than 1000 will fail with "first" must be less than or equal to 1000. Backward pagination with last/before is no longer supported on these fields.

This change protects the server from unbounded queries that could cause excessive memory usage and slow response times.