Skip to main content

Overview

The Phoenix CLI now supports datasets, experiments, and annotations. You can pull evaluation data, experiment results, and human feedback directly into your terminal or pipe them into AI coding assistants.
# List available datasets
px datasets

# Fetch dataset examples
px dataset <dataset-name>

# Export experiment results
px experiments --dataset <dataset-name> ./experiments/

# Get experiment details
px experiment <experiment-id> --format json

# Include annotations in traces
px traces --limit 10 --include-annotations

Common Workflows

Analyze Dataset Examples

Pull test cases for review:
# Get all examples from a dataset
px dataset <dataset-name> --format raw --no-progress | \
  jq '.examples[] | {input, output}'

# Filter by split (if dataset has splits defined)
px dataset <dataset-name> --split train --format raw --no-progress | \
  jq '.examples[]'

Review Experiment Results

Check evaluation results:
# Get experiment runs with their inputs and outputs
px experiment <experiment-id> --format raw --no-progress | \
  jq '.[] | {input, output, latency_ms, trace_id}'

# Find runs with annotations
px experiment <experiment-id> --format raw --no-progress | \
  jq '.[] | select(.annotations | length > 0)'

Use Annotations

Access human feedback on traces:
# Get traces with annotations
px traces --include-annotations --limit 50 --format raw | \
  jq '.[] | select(.spans[].annotations != null)'

Export for Analysis

Save data to files for further processing:
# Export traces to directory
px traces --include-annotations --limit 100 ./production-samples/

# Export experiments to directory
px experiments --dataset <dataset-name> ./all-experiments/

# Filter examples with jq
px traces --include-annotations --format raw --no-progress | \
  jq '.[] | select(.spans[].annotations[] | select(.name == "quality" and .score >= 4))'

Dataset and Experiment Commands

Datasets

# List all datasets with metadata
px datasets --format json

# Fetch dataset with specific splits
px dataset <dataset-name> --split train --split validation

# Get specific version
px dataset <dataset-name> --version <version-id>

Experiments

# List experiments for a dataset
px experiments --dataset <dataset-name>

# Export all experiment data
px experiments --dataset <dataset-name> ./experiments/

# Fetch single experiment with full details
px experiment <experiment-id> --format raw

Annotations

# Include annotations in any trace command
px traces --include-annotations --limit 10

# Get specific trace with annotations
px trace <trace-id> --include-annotations

Example: Reviewing Experiment Results

Get data from an experiment and dataset:
# Export experiment runs to a file
px experiment <experiment-id> --format raw --no-progress > experiment-runs.json

# Extract inputs and outputs
jq '.[] | {input, output, latency_ms}' experiment-runs.json > run-summary.json

# Export dataset examples
px dataset <dataset-name> --format raw --no-progress | \
  jq '.examples[] | {input, output}' \
  > dataset-examples.json
You can then ask an AI assistant to review the experiment results or analyze patterns in the dataset.

Installation

npm install -g @arizeai/phoenix-cli
Or run directly:
npx @arizeai/phoenix-cli datasets

Upgrading

If you already have the CLI installed, upgrade to 0.2.0 or later:
npm update -g @arizeai/phoenix-cli
Check your version:
px --version

Feedback

Share feedback or contribute on GitHub.