Your first agent

Five steps, no code. The slow part is gathering the documents.

1. Category and identity

You choose the name, area, AI model and who uses it: public or only people with an account (corporate agent). The category sets the starting prompt and the suggested model: compliance, collections and after-sales start with the most capable one; help desk and appointments, with Sonnet (catalog).

Everything else it can do starts off (except generating documents in technology agents) and is changed later, in What it can do → Capabilities (and What it knows → Memory):

Option	What it allows
Web search	Searching the internet when answering
Generate documents	Reading uploaded documents and images and creating downloadable files (Word, text, code or HTML)
Schedule tasks	Leaving tasks for later ("remind me on Monday"); see Scheduled tasks
Remember people	Saving notes about each person between conversations
Learn its own knowledge	Proposing notes for everyone; an administrator approves them before use

Turn on the last two only if your policy allows it.

2. Role

Describe how it should behave, or ask the platform to generate it from the role, tone and rules. Each change is a version you can compare or revert.

3. Knowledge

PDF, Word (.docx), Excel (.xlsx), HTML, XML, Markdown or text, up to 20 MB per file; also a CSV with several articles or a ZIP.

A scanned PDF doesn't work

It is an image with no text; the platform detects it and warns you.

4. Connections

Optional: Connect your systems.

5. Review and deploy

Deploy agent doesn't publish it: in the Where it answers tab, generate the public link or copy the widget and authorize your domain (Widget).

Before showing it, test it with real questions, including one whose answer is not in the documents: it should say it doesn't know.