LEARN · BUILD · EXPERIMENT

AI Lab

One idea at a time. Real code you can change.

0 / 10

missions checked

Your learning path

beginner1 of 10

Learn by doing4 min

01 / Code that does something

Make your first tool call

Start with a workflow: a tool supplies data, and your code follows a fixed rule.

YOU WILL BUILD

Return the library for 18 students, and explain how the same rule handles a different group size.

Follow the picture, try a prediction, then open the editor. You can revisit any step.

Read a little more

Read before you recommend

Calling a tool runs an operation. Here it reads fictional room records. This first program follows your fixed instructions; the model-driven decision loop comes in mission 3.

tools.listRooms() → a list of room objects

Turn the input into a condition

The same function should work when the club grows. Compare capacity with task.students rather than writing one answer into the program.

room.capacity >= task.students

The editor opens with a starter and explained snippets.

See the ideaIllustrated example · 18 students

STEP 1 OF 3

Read the input

The task supplies the group size. A tool reads the classroom room records; your code uses its returned objects.

task.students = 18
tools.listRooms() → studio, library, hall

Try a group size 18 students

Design studio

12 seats

Library room

24 seats

School hall

40 seats

find → "library"

This preview checks capacity only, in list order. You will add price and availability in the next lesson. It is not a run of your editor code.

1 / 3

The school records and model decisions in these walkthroughs are authored simulations. Your JavaScript runs when you choose Run.

Why we teach it this way

The goal is to seat a school robotics club. A room name that sounds convincing is not enough: the capacity must fit the actual number of students.

AnthropicBuilding effective agentsEngineering guidance on simple workflows, tool-result feedback, stopping limits, and human checkpoints. The original article is from December 2024; its tooling landscape has changed.MDN Web DocsJavaScript GuideLanguage reference for functions, objects, control flow, and asynchronous JavaScript. It does not claim that ordinary JavaScript rules are a language model.

Explore real agent tools Optional · local setup

OPEN-SOURCE PATH · OPTIONAL

Take your agent local

Use LangGraph, a real room tool, and a local Qwen model through Ollama. Start with the offline test; connect the model when you are ready.

Optional · Node.js 22+ · Your own computer · Ollama model download for live runs

GUIDED SETUP · 1 OF 4

Create your project

Download the setup file into a new folder. Run the commands below, then open the generated README in your editor.

The setup file creates seven readable project files in a new folder. It never installs packages automatically or overwrites an existing project.

First, run the offline checks

node setup-ai-lab.mjs
cd stemtiq-local-agent-lab
npm install
npm run check
npm test

Inspect the project in this IDE

These files run on your computer. The browser’s Run button executes agent.js in the classroom simulator. Install Ollama and download the local model only when you choose to follow the README.

One read-only tool · Up to 4 model turns / 3 tool calls · 30-second run abort · No booking or shell tools

Official documentation and checks

Pinned packages, TypeScript, offline graph tests, and the missing-service message were checked. Live model output has not been verified.

LangGraph Graph API quickstart ChatOllama integration Ollama tool calling Local Ollama setup and cloud controls Qwen3:4b model and license

Explore the separate OpenAI SDK example Requires an API key in your local environment

THE NEXT STEP

Build with the OpenAI Agents SDK.

Explore a separate example that uses a real agent SDK. Read the code here, then run it in your own environment with an API key. This browser lab never asks for one.

Agent + run: Configure the real model and let the SDK manage the tool loop.
tool + Zod: Define an executable tool and validate its inputs.
maxTurns: Bound model turns; still check outputs and permissions.

Local project setup

# Use Node.js 22 or later in a separate folder.
# Real runs require an OpenAI API account/key and may incur API charges.
npm init -y
npm pkg set type=module
npm install @openai/agents@0.18.0 zod@4
npm install --save-dev typescript tsx @types/node@22
# Save the downloaded example as real-agent.ts.
# Set OPENAI_API_KEY privately in your terminal environment before running.
# Optional: set OPENAI_MODEL to a compatible model available to your API project.
npx tsx real-agent.ts

AI Lab

18 studentsNot run

Take your agent local

Create your project

Build with the OpenAI Agents SDK.

Not run