AI-powered analysis of your Phoenix observability data using the "bash + files" paradigm.

Phoenix Insight materializes your traces, experiments, datasets, and prompts as a structured filesystem, then uses an AI agent to explore and analyze the data with standard Unix commands you already know: cat, grep, jq, awk, and more.

Traditional AI observability tools hide data behind opaque APIs and dashboards. Phoenix Insight takes a different approach:

• Transparent: Every query the agent runs is visible. You see the exact commands, the files it reads, and the data it processes.
• Reproducible: Commands are just bash. Copy them, modify them, run them yourself. Debug AI-driven analysis like you debug any script.
• Extensible: The data is just files. Use any tool in your Unix toolkit: ripgrep for faster search, custom scripts for specialized analysis, or pipe output through your favorite tools.

• Node.js v22 or newer
• Anthropic API key

Set your API key before running:

export ANTHROPIC_API_KEY=sk-ant-api03-...

Get an API key at console.anthropic.com.

Phoenix Insight analyzes data from Phoenix, an open-source observability platform for LLM applications. Choose one:

Option A: Phoenix Cloud (Recommended)

1. Create a free account at app.phoenix.arize.com
2. Note your space name from the URL (e.g., https://app.phoenix.arize.com/s/my-space)
3. Generate an API key in Settings → API Keys

Option B: Self-Hosted Phoenix

Run Phoenix locally with Docker:

docker run --pull=always -d --name arize-phoenix -p 6006:6006 arizephoenix/phoenix:latest

Phoenix will be available at http://localhost:6006.

npm install -g @cephalization/phoenix-insight

Run the interactive setup:

phoenix-insight init

When prompted:

• Phoenix URL: Enter https://app.phoenix.arize.com/s/<your-space> for Cloud, or press Enter for http://localhost:6006 (self-hosted default)
• API Key: Enter your Phoenix Cloud API key, or press Enter to skip (not needed for self-hosted)

Send a test trace to Phoenix:

phoenix-insight seed

This sends a traced "Hello, world!" message to your Phoenix instance. If successful, you'll see a link to view the trace in Phoenix.

You're ready to use Phoenix Insight! Try a query:

phoenix-insight "What traces do I have?"

npm install -g @cephalization/phoenix-insight

Or run directly with npx:

npx @cephalization/phoenix-insight "your query"

# Analyze error patterns
phoenix-insight "What are the most common errors in my traces?"

# Find performance bottlenecks
phoenix-insight "Which endpoints have the highest latency?"

# Explore LLM usage
phoenix-insight "Show token usage breakdown by model"

Start an interactive session for multiple queries:

phoenix-insight

phoenix> What projects have the most errors?
[Agent analyzes and responds...]

phoenix> Show me the top 5 slowest traces
[Agent investigates...]

phoenix> exit

Launch a visual interface with chat and structured reports:

phoenix-insight ui

This opens a browser with:

• Chat panel for natural language queries
• Report panel for structured analysis results
• Session history and report persistence

• CLI Documentation - Full command reference, configuration options, and advanced usage
• Development Guide - Monorepo setup, architecture, and contributing

Apache-2.0 - See LICENSE for details.