Metric Charts Above Data Tables#

July 7, 2026

Available in arize-phoenix 17.20.0+

Keep the numbers that matter in view while you work through your data. A resizable strip of metric charts now sits above the spans, traces, and sessions tables. Pick the charts you want per table from a catalog, and Phoenix remembers your selection per project.

• Per-table selection — choose different charts for the spans, traces, and sessions views
• Interactive — filter series from the legend and drag across a chart to zoom the time range
• Persistent — your chart choices are saved per project

Trace Tree Search#

July 6, 2026

Available in arize-phoenix 17.19.0+

Finding a span in a deep, deeply nested trace no longer means scrolling. The trace detail view now has a search box that filters the trace tree as you type, so you can jump straight to the span you care about.

Dataset Labels and Annotation Config Assignment over REST#

July 4, 2026

Available in arize-phoenix 17.16.0+ (server), arize-phoenix-client 2.12.0+ (Python, typed access)

Two organizational workflows that were previously UI-only are now scriptable over the REST API.

Dataset labels can be created, listed, updated, deleted, and applied to datasets:

# Create a labelcurl -X POST "$PHOENIX_HOST/v1/dataset_labels" \  -H "Authorization: Bearer $PHOENIX_API_KEY" \  -H "Content-Type: application/json" \  -d '{"name": "golden", "color": "#00cc88", "description": "Curated eval sets"}'# Apply a label to a dataset (idempotent)curl -X PUT "$PHOENIX_HOST/v1/datasets/$DATASET_ID/labels/$LABEL_ID" \  -H "Authorization: Bearer $PHOENIX_API_KEY"# List the labels applied to a datasetcurl "$PHOENIX_HOST/v1/datasets/$DATASET_ID/labels" \  -H "Authorization: Bearer $PHOENIX_API_KEY"

Annotation configs can be assigned to and removed from projects, and both endpoints accept either an ID or a name for the project and config:

# Assign an annotation config to a project (idempotent)curl -X PUT "$PHOENIX_HOST/v1/projects/my-project/annotation_configs/response-quality" \  -H "Authorization: Bearer $PHOENIX_API_KEY"# Replace the full set of configs on a projectcurl -X PUT "$PHOENIX_HOST/v1/projects/my-project/annotation_configs" \  -H "Authorization: Bearer $PHOENIX_API_KEY" \  -H "Content-Type: application/json" \  -d '{"annotation_config_ids": ["'"$CONFIG_ID"'"]}'

Filter Projects by Name#

July 4, 2026

Available in arize-phoenix 17.16.0+

GET /v1/projects now accepts a name_contains query parameter for a case-insensitive substring match, so you can locate projects without paging through the full list.

curl "$PHOENIX_HOST/v1/projects?name_contains=eval" \  -H "Authorization: Bearer $PHOENIX_API_KEY"

Baseline Experiments#

July 4, 2026

Available in arize-phoenix 17.16.0+

Mark one experiment per project as the baseline to make it the reference point in comparison workflows. Set or clear the baseline from the experiment action menu — the baseline is flagged with a badge across experiment tables and compare views, and the choice persists across sessions.

Annotation Config CLI Commands#

July 4, 2026

Available in arize-phoenix-client 2.12.0+

The px CLI can now manage annotation configurations directly, making it easy to script annotation setup or drive it from a coding agent.

# List all annotation configurationspx annotation-config list# Fetch one by name or IDpx annotation-config get response-quality# Create a categorical pass/fail ratingpx annotation-config create \  --type CATEGORICAL \  --name response-quality \  --value good=1 \  --value bad=0 \  --optimization-direction MAXIMIZE# Update just the descriptionpx annotation-config update response-quality \  --description "Pass/fail rating from human review"

Classification Metric Evaluators in TypeScript#

July 4, 2026

Available in @arizeai/phoenix-evals 1.1.0+ (TypeScript)

The TypeScript evals package now ships built-in code (non-LLM) evaluators for precision, recall, and F-score, mirroring the Python PrecisionRecallFScore evaluator. They support binary classification via positiveLabel and multi-class classification with macro, micro, or weighted averaging.

import {  createPrecisionEvaluator,  createPrecisionRecallFScoreEvaluators,} from "@arizeai/phoenix-evals/code";// Binary classificationconst precision = createPrecisionEvaluator({ positiveLabel: "spam" });const p = await precision.evaluate({  expected: ["spam", "ham", "spam"],  output: ["spam", "spam", "ham"],});// Multi-class, weighted average — precision, recall, and F-score togetherconst { precision: mp, recall, fScore } = createPrecisionRecallFScoreEvaluators({  average: "weighted",});const f = await fScore.evaluate({  expected: ["cat", "dog", "cat", "bird"],  output: ["cat", "cat", "cat", "bird"],});

For a one-shot computation without constructing an evaluator, use computePrecisionRecallFScore from the same import path.

Log Spans from the TypeScript Client#

July 2, 2026

Available in @arizeai/phoenix-client 6.12.0+ (TypeScript)

The TypeScript client gains logSpans, mirroring the Python client’s log_spans. It submits spans directly to a project using Phoenix’s simplified span structure — the same shape returned by getSpans — with no OpenTelemetry setup required.

import { logSpans } from "@arizeai/phoenix-client/spans";const result = await logSpans({  project: { projectName: "my-project" },  spans: [    {      name: "my-span",      context: { trace_id: "...", span_id: "..." },      span_kind: "CHAIN",      start_time: "2026-07-02T00:00:00Z",      end_time: "2026-07-02T00:00:01Z",      status_code: "OK",    },  ],});console.log(`Queued ${result.totalQueued} of ${result.totalReceived} spans`);

Invalid or duplicate spans raise a SpanCreationError with invalidSpans and duplicateSpans details so you can see exactly which spans were rejected.

Claude Sonnet 5 in the Playground#

July 1, 2026

Available in arize-phoenix 17.15.0+

Claude Sonnet 5 is now selectable in the Playground and prompts for both the Anthropic (claude-sonnet-5) and AWS Bedrock (anthropic.claude-sonnet-5) providers.

Table and Navigation Improvements#

July 4–6, 2026

Available in arize-phoenix 17.16.0+ through 17.19.0+

A batch of workflow upgrades landed across the data tables and side navigation:

• Error column on the spans and traces tables surfaces the status message recorded when a span errors (arize-phoenix 17.16.0+)
• Prompt version columns — the prompts table now shows the version count, latest version, and version tags with links (arize-phoenix 17.16.0+)
• Shift-click range selection — select a contiguous range of rows with shift-click across the examples, experiments, spans, traces, and annotation-config tables (arize-phoenix 17.17.0+)
• Recent searches — the span filter typeahead shows your last few filter conditions per project (arize-phoenix 17.17.0+)
• Consolidated account menu — profile, docs, support, management console, log out, and the light/dark/system theme toggle now live in a single menu at the bottom of the side nav (arize-phoenix 17.19.0+)