Your first 30 minutes

In one line. One guided sit-down: build a small but real Agentic Search assistant over three documents, and meet every core idea on the way. You’ll be able to. Create a collection, hire an agent, ground it in your own documents, test it, and hand it to a colleague — in half an hour. You’ll need. A project you can build in, and three PDFs you know well — policy documents, product guides, anything your colleagues ask questions about.

Don’t read this chapter. Do it. Every concept you meet here — collection, agent, knowledge attachment, playground — carries the rest of the academy.

Minutes 0–5 · Look around

Sign in and you land on the Command Deck — the cross-workspace home (press Ctrl K to jump anywhere). Open your project from Jump back in or Browse everything, and you’re on the surface your users get: the workspace, chat, the pages people work from.

Now open Studio from the icon in the top bar — hover it and the tooltip reads Go to Studio Menu. Studio is where builders work, and your users never see it. If the icon isn’t there, either the project has Studio switched off or your access doesn’t allow building — ask your administrator.

The top bar changes once you’re inside. Home and the four Studio pillars — Agents · Data · Experience · Governance — appear across the top, and a Hubs chip and an Exit Studio button appear on the right. Those four names are the whole map: agents and what they’re made of, data and how it arrives, the experience your users see, and who’s allowed to do what. Take the two-minute tour in E1 · Getting oriented if you want names for everything, then come back.

Minutes 5–10 · Create a Knowledge collection

Studio ▸ Data ▸ Collections ▸ + Collection

Name it something honest — HR Policies, Product Guides.
Under Purpose, choose Knowledge. That one choice is what makes this an Agentic Search build: every document is indexed for your agents to read as soon as it lands. No lifecycle, no stages.
Upload your three PDFs.

Indexing takes a moment — a big upload can take a few minutes. Open the collection’s Index Health tab to watch it finish; it also lists anything that failed, so you can retry it there.

While they index, understand what you just did: you gave your future agent its ground truth. An agent answers only from what you attach, and that boundary is what makes its answers trustworthy.

Minutes 10–20 · Hire your first agent

Studio ▸ Agents ▸ Agents ▸ + Agent

Agents is the name of both the pillar and its first page, so the full path reads Studio ▸ Agents ▸ Agent Toolkit ▸ Agents. Either form gets you to the same place — the Agents page. + Agent offers three kinds — Agent, XFlow and Mesh; choose Agent.

Persona — describe the worker in plain language: “You answer employee questions about our policies. Answer only from the attached documents, and always cite your sources. If the documents don’t contain the answer, say so.”
Model — pick one of the language models registered for this project, or leave it on the project default. (Someone registers those models, and marks one the default, under Studio ▸ Agents ▸ Language Models.)
Knowledge — attach HR Policies. This is the load-bearing wire of the whole build: it is what turns a generic chatbot into your expert.

Save. That’s a hire.

Minutes 20–25 · Ask it things

Open the agent’s Playground and ask real questions — the ones people actually bring to your team. Watch what comes back: grounded answers with citations you can click through to the exact source passage.

Ask something the documents don’t cover. A well-instructed agent says it doesn’t know instead of inventing — if yours doesn’t, tighten the persona line about refusing and try again. This test-refine loop in the playground is how every agent on the platform gets good.

Minutes 25–30 · Give it to someone

Your users don’t open Studio — they open Chat. Check the chat channel reads the way you want under Studio ▸ Experience ▸ Chat, then leave Studio with Exit Studio, copy the address you land on, and send it to one colleague with the message “ask it anything about our policies.”

They need access to the project to open it. If they aren’t in it yet, add them under Studio ▸ Governance ▸ Users & Access. Then watch what they ask. Their third question is usually the moment this stops being a demo.

What you just built

Agentic Search — solution shape 1 of 3. A Knowledge collection, one agent, one attachment, one channel your colleagues can use. The other two shapes are the same two heroes doing bigger jobs:

Add a schema, a lifecycle, and a review queue and the agent stops answering questions and starts processing work — that’s an Agentic Workflow.
Add more collections, more agents, and governance across all of it — that’s an Agentic System.

Where to go next

Make this build production-worthy: the Search pattern, then the full worked build in B1 · Policy assistant.
Meet the agent editor properly: A3 · Your first agent.