# Agent (/docs/en/guides/agent)



The Agent reads your practice's records and drafts notes, letters and templates. It can also propose bookings, emails, forms and tasks. It works with your permissions and your access to each patient, and it needs the **Use AI features** permission, which every default role has.

The Agent can make mistakes, so check its work before you keep or confirm anything. Diagnosis, treatment and dose stay with you.

## Ask the Agent [#ask]

Open **Agent** in the sidebar, or choose the **Agent** button in a note, letter, form or task to work beside it. Beside a document, the Agent already knows what you have open. On the Agent page, name what you mean.

* Type **@** to mention a patient or a document.
* Type **/** to see the **Commands**: your daily brief, and getting ready for a patient visit.
* On a new chat, use a starter such as **Brief me on today** or the one that goes through your open tasks.

Say what you want, which source to use and what to leave out. Check the patient's name in your request and in the context before you send it.

Open **History** to return to earlier chats. Documents the Agent writes open beside the conversation.

If a patient is marked **Exclude from AI assistance** in their file, the Agent cannot use their record. The patient is shown as **Excluded from AI**, and a prompt cannot change that.

## Review the changes it proposes [#review]

The Agent never changes a record on its own.

* In a record or in patient details, each proposed change is a row you can **Keep** or **Dismiss**. Nothing joins the record until you keep it, and **Keep all** applies to the whole list.
* In a visit note or letter, proposed text appears in the document with **Keep** and **Dismiss**. Kept clinical items are drafts until you add them to the record, as described in [Visits and notes](/docs/en/guides/visits).
* Templates and forms the Agent writes are saved as unpublished drafts. You publish them yourself.

Check dates, values and units against the source. If a run is interrupted, look at the document and its pending changes before you ask again. **Restore proposal** brings back a saved proposal.

## Confirm what the Agent wants to do [#approve]

For actions that reach people or change your calendar, the Agent shows a card marked **Needs your go-ahead**. Read the card, then choose the action button (for example **Send** or **Book**) or **Decline**. Declining stops that action.

The Agent always asks before it:

* sends an email or a form to a patient
* books, moves or cancels an appointment
* moves a file to another shelf

<Callout type="warn">
  Confirming **Send** emails the patient. The card shows the recipient and the message. After you confirm, the email waits 30 seconds, and you can use **Undo** during that time. Emails go only to the patient's address on file.
</Callout>

A card for a booking, move or cancellation also says whether the patient will be emailed.

For tasks and file renames, you choose whether the Agent asks first. Open **Agent settings → Approvals** and switch each action on or off. Creating a task is on by default, and commenting on a task is off. See [Agent settings](/docs/en/guides/agent-settings).

A draft is not a sent email and a proposal is not a booking. Check the result in the patient's **Emails** page or the calendar.

## AI credits [#credits]

Every reply uses AI credits. Your practice shares a pool of credits that renews each month. Owners and admins can give a person a monthly limit.

The credit gauge by the message box shows how many credits are left this month, what this chat has used and what the last reply cost. Choose **Usage details** for more. **Agent settings → Usage** shows your own figures.

When your limit or the practice's credits run out, the Agent stops answering until the credits renew, and a message shows the renewal date. An owner can turn on **Extra credits** to keep the Agent working. Plans and prices are in [Billing](/docs/en/guides/billing).

## Related [#related]

* [Skills](/docs/en/guides/skills): teach the Agent your own procedures.
* [Agent settings](/docs/en/guides/agent-settings): approvals, languages and usage.
* [Agent and you](/docs/en/get-started/agent-and-you)
* [Agent problems](/docs/en/troubleshooting/agent-problems)