# Choose what each AI agent can read

> Pin Library items, search portals, or run an agent without any knowledge.

Each agent reads only the knowledge you allow. Getting this right keeps internal material away from customers and keeps answers focused.

## Library

On the agent's edit page, **Knowledge** → **Library** works in three ways:

| Setting | What the agent reads |
| --- | --- |
| Items chosen with **Pick items** | Only those items |
| Nothing picked | Every ready Library item, including ones you add later |
| **Don't use the Library** on | No Library data at all |

Select **Pick items** to open **Select Library for Agent**. It lists only items that are **Ready**. Choose the ones you want and select **Use selected Library**. Turning **Don't use the Library** off again brings back the items you had picked.

Tags on Library items don't affect what an agent reads. Only this selection does.

## Portals

Agents can also search the published articles of your portals, so an answer can come from a Help Center article.

| Setting | What the agent searches |
| --- | --- |
| Portals selected | Only those portals |
| No portals selected (**All portals**) | Every enabled portal |

## Recommended setups

- **Customer-facing agent:** select only public material, such as your website and public help articles.
- **Internal team agent:** add playbooks and internal processes, and keep it off public deployments.
- **Narrow agent,** such as one that only books appointments: select just the items it needs.

## In AI Chat

When you work in **AI Chat**, select **Library** in the composer to attach items to that conversation, without changing any agent.

## Check the result

After changing access, open **AI Agents** and choose **Chat** → **New conversation** on the agent's row. Ask a question only the right material can answer.

<Tip>
Review a Library item before it reaches a customer-facing agent. If it contains internal notes, leave it out of that agent's selection.
</Tip>