Arize 101 Workshop — Local Setup

Run the AI Engineer World's Fair — Arize 101 Workshop notebook locally (it was authored for Google Colab; this repo adapts it for a local venv).

Prerequisites

• Python 3.12 (via pyenv) and uv
• Node.js + the claude CLI on your PATH — the Claude Agent SDK shells out to it
• An Anthropic API key and an Arize AX account (API key + Space ID)

1. Install

cd ArizeEvals
uv venv --python 3.12 .venv
uv pip install --python .venv -r requirements.txt
.venv/bin/python -m ipykernel install --user --name arize-evals --display-name "Arize Evals"

2. Add your secrets

cp .env.example .env
# edit .env and fill in ANTHROPIC_API_KEY, ARIZE_API_KEY, ARIZE_SPACE_ID

.env is gitignored. The notebook loads it automatically (cell under "Step 1: Set Up Tracing") via python-dotenv.

3. Run

.venv/bin/jupyter lab "AI_Engineer_World's_Fair_Arize_101_Workshop.ipynb"

Pick the "Arize Evals" kernel, then run cells top to bottom.

Heads up: ANTHROPIC_BASE_URL

If your shell has ANTHROPIC_BASE_URL set (a custom gateway/proxy), the notebook's agent may fail: it uses WebSearch, a server-side Anthropic tool that requires api.anthropic.com directly. If calls fail or web search returns nothing, launch Jupyter with the proxy disabled:

env -u ANTHROPIC_BASE_URL .venv/bin/jupyter lab "AI_Engineer_World's_Fair_Arize_101_Workshop.ipynb"

(or leave ANTHROPIC_BASE_URL commented out in .env).

What changed from the Colab original

Only the secrets cell. Colab's from google.colab import userdata was replaced with a python-dotenv loader that reads the three keys from .env. Everything else runs unmodified.