Pip Tutorial

Build Your First AI Agent

A rubber duck that checks in with you every morning, remembers what you tell it, and gets smarter over time.

The to-do app taught a generation of developers how web frameworks worked. This tutorial is the equivalent for AI agents.

You're going to build Pip — a personal AI agent that sends you a daily check-in email, listens to your reply, and builds a memory of who you are. By the end, your duck will notice patterns in your life that you haven't noticed yourself.

If you've read the guide, you know Pip as the rubber duck who's furious about being replaced by AI. In this tutorial, you're going to build the thing he was afraid of — and in doing so, maybe prove him wrong. A good companion doesn't replace the old one. It just shows up differently.

What you'll build

An AWS-powered agent with a sense → reason → draft → notify → learn loop:

Sense: A scheduled trigger fires every morning
Reason: Claude (via Amazon Bedrock) reviews your memory profile and decides what to ask you
Draft: It composes a personalized check-in question
Notify: You get an email. You reply.
Learn: Your reply is parsed for new memories, which inform tomorrow's question

The stack

SST (Ion) — Infrastructure as code, live Lambda development
AWS Lambda — The agent's brain runs here
Amazon Bedrock — Claude, no API keys to manage
DynamoDB — Where memories live
SES — Sends and receives email
S3 — Landing zone for inbound email
EventBridge — The schedule that makes it an agent

Cost

Running this for yourself costs $1–3/month. Lambda, DynamoDB, S3, and EventBridge are effectively free at this scale. SES is a penny. Bedrock (Claude Haiku) is the only real cost — about 60 calls/month at ~1K tokens each.

The three parts

Part 1: The Duck Says Hello

Deploy your first Lambda. Get an email from your duck. Reply to it. The duck acknowledges — but forgets you by morning. Every day is a first date.

Part 2: The Duck Remembers

Give the duck a memory. It extracts what matters from your replies, stores it, and uses it to ask better questions. After a week, the first-date problem is gone.

Part 3: The Duck Knows You

The duck starts seeing patterns. A weekly synthesis finds threads across your memories and surfaces insights you haven't noticed. The moment it says something true about you that you didn't tell it — that's when it stops being a tutorial project.

Prerequisites

An AWS account — create one here if you don't have one
Node.js 18+ — node --version to check
AWS CLI configured — aws sts get-caller-identity should return your account
Bedrock model access — enable Claude Haiku in the Bedrock console
Basic terminal comfort — you can run npm install and read error messages

You don't need to know TypeScript deeply. You don't need to have read the guide. You just need an hour and a reason to build something.

Reference code

Want to skip ahead or get unstuck? The complete reference implementation is on GitHub. You can clone it and deploy directly, or use it as a reference while you follow the tutorial.