01.17.2026 Phoenix CLI: Terminal Access for AI Coding Assistants

CLI Reference

Full documentation

npm

Install from npm

Overview

@arizeai/phoenix-cli is a command-line interface for retrieving trace data from Phoenix. It provides the same observability data available in the Phoenix UI, accessible through shell commands and file exports.

# List available projects
px projects

# Fetch recent traces
px traces --limit 10 --format json

# Export traces to files
px traces ./debug-data --limit 50

Why a CLI Matters Now

AI coding assistants—Claude Code, Cursor, Windsurf, Codex, Gemini CLI—operate through two primary interfaces: the terminal and the filesystem. They run shell commands, read output, and process files. This is how they interact with your development environment. For these assistants to help debug and improve AI applications, they need access to observability data through these same interfaces. A browser-based UI, while useful for humans, is inaccessible to an AI assistant working in your IDE. The Phoenix CLI bridges this gap. Your AI assistant can now:

Query your Phoenix instance directly via shell commands
Retrieve trace data as structured JSON
Export traces to files for analysis
Pipe output to tools like jq for filtering

Practical Workflows

Debugging with AI Assistance

When an agent fails, you no longer need to manually copy trace data from the UI. Instead:

Fetch the last 5 traces from my project using px and identify why 
the agent is failing on tool calls.

Your AI assistant runs px traces --limit 5 --format raw --no-progress, parses the JSON output, and analyzes the span data directly.

Exporting for Analysis

Build evaluation datasets by exporting traces to disk:

px traces ./eval-dataset --limit 100 --last-n-minutes 60

Each trace is saved as a separate JSON file, ready for processing by scripts or AI assistants.

Filtering and Processing

Use standard Unix tools alongside the CLI:

# Find error traces
px traces --limit 50 --format raw --no-progress | jq '.[] | select(.status == "ERROR")'

# Extract model names from LLM spans
px traces --limit 20 --format raw --no-progress | \
  jq -r '.[].spans[] | select(.span_kind == "LLM") | .attributes["llm.model_name"]' | sort -u

Output Formats

pretty — Human-readable tree view showing span hierarchy
json — Formatted JSON with indentation
raw — Compact JSON for piping to other tools

Installation

npm install -g @arizeai/phoenix-cli

Or run directly without installation:

npx @arizeai/phoenix-cli traces --limit 1

Looking Ahead

As AI assistants take on larger roles in software development, they will need access to the same data and tools that human developers use. Observability platforms built exclusively around browser UIs will become limiting. The CLI is our response to this shift. It makes Phoenix data available where AI assistants already work—in the terminal and through files. We expect this pattern to become standard as the tooling ecosystem adapts to AI-assisted development. Feedback and contributions are welcome on GitHub.

CLI Reference

npm

​Overview

​Why a CLI Matters Now

​Practical Workflows

​Debugging with AI Assistance

​Exporting for Analysis

​Filtering and Processing

​Output Formats

​Installation

​Looking Ahead