Build Your AI Assistant with Claude

6 Interactive Sessions

Short, interactive sessions — watch it work, steer it, then build it yourself. Go deeper anytime with the full code walkthrough.

1. 1

   Talking to Claude — the messages API

   You want an assistant that summarizes your unread email, checks your 2pm, and drafts a prep doc. It starts with one small thing: a program that can TALK to Claude — send a message, get a streamed response back. That's the API.

2. 2

   Giving Claude Hands — the tool-use loop

   Ask a plain chat model "what's on my calendar at 2pm?" and it will confidently make something up — it can't see your calendar. Tools fix that: you give Claude a function it can CALL, your code runs it, and hands back the real answer.

3. 3

   Email & Calendar — the assistant's first real hands

   Now the assistant gets real hands: read your unread email, check your calendar, draft a reply. "Summarize my unread and check my 2pm" becomes real — and one clear line keeps it safe: it PREPARES, you approve anything that can't be undone.

4. 4

   Look It Up — a research agent with sources

   "Draft a prep doc for my 2pm with Acme" needs facts the assistant simply doesn't have — who Acme is, what's new. So it researches: search, read the top results, and synthesize a briefing where every claim carries its source.

5. 5

   Remember Me — conversation history & preferences

   You told it "I prefer morning meetings" yesterday; today it forgot. Without memory, every conversation starts from zero. Memory is two things: the history of this chat, plus a small profile of you that persists — so the assistant stops re-introducing itself.

6. 6

   Run It on Your Machine — a private local CLI

   An assistant is only useful if it runs when you need it. So you package everything you built into a small CLI on your own machine — your API key, your connected accounts, your data staying local — and type "summarize my morning" from the terminal.

What you walk away with