> For the complete documentation index, see [llms.txt](https://docs.openiam.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.openiam.com/end-user-guide-for-selfservice/ai-assistant-conversations.md).

# Your AI assistant conversations

Use the AI assistant in SelfService — type and send a question, ask about your access, roles and colleagues, ask about the requests you made, read its cards and tables, request access for yourself, pi

**Applies to:** OpenIAM 2026.11.2

The **AI Assistant** in SelfService answers questions about your access, your requests and your approvals. It keeps every conversation you have with it, so you can come back to one later and carry on where you left off.

Use this page to find your way around the assistant, to return to something you asked it earlier, or to start again with a clean conversation.

## Opening the assistant

Click **Assistant** at the top of the SelfService menu, above the other menu groups. It has a sparkle icon and a **Beta** label.

Your organization decides who gets the assistant. If you do not see **Assistant** in your menu, it has not been turned on for you, or it is not available at the moment.

While the assistant is out of service, SelfService hides the **Assistant** menu entry. You do not need to reload the page. The entry comes back by itself, usually within a minute of the assistant working again. If the assistant could not be reached at all, it comes back the next time you open a page once it works again.

A conversation you already have open stays open, and so does your list of conversations. A question you send while the assistant is out of service shows a box that says what went wrong, as described in [When the assistant cannot answer](#when-the-assistant-cannot-answer). Click **Retry** once the assistant is back.

If you open the assistant from a bookmark or a link while it is not available, you see `The assistant is unavailable` and the reason:

* `The assistant has not been enabled for your organisation. Contact your administrator.` — you do not have the assistant. Ask your IT administrator if you need it.
* `The assistant is temporarily degraded. We're working on it, please try again later.` — the assistant is having a problem. Try again later.

### If the assistant is taken away while you use it

Your organization can turn the assistant off, or take it away from you, while you have it open. This does not happen at once: depending on how your administrator made the change, it takes from under a minute to about an hour. After that, nothing you send reaches the assistant any more. The next time you send a question, or open, rename, pin, archive, unarchive or delete a conversation, the assistant shows `The assistant is unavailable` and the reason, as above. From then on, the **Assistant** menu entry stays hidden for the rest of your visit. When you reload the page, SelfService checks again. You cannot open or change your earlier conversations until you have the assistant again.

If the assistant cannot check whether you have it, it does not answer and shows an error instead. Click **Retry** to ask again a moment later.

### Before your first question

The first time you send a question, a `Before you start` box opens. It explains:

* that the assistant only sees what you are allowed to see in OpenIAM, such as your own details, approvals, requests and team;
* which model the assistant uses, and that your questions, and the information needed to answer them, are sent to the company that provides the model, outside your organization's network;
* that the assistant never changes anything on its own, and always asks you to confirm an action first;
* how long your conversations are kept.

Click **Got it**. The question you typed is then sent for you, so you do not need to type it again. The box does not appear again in that browser.

To read it again later, click **More options** (the three dots at the top right of the assistant), then **Disclaimer & privacy**. You can also click **Disclaimer** under the message box. If you use another browser or another computer, or clear your browser's data, the box appears once more before your first question there.

## Finding your way around the assistant

The top of the assistant shows its title, **Assistant**, with a **Beta** label beside it. The label means the assistant is new: check what it tells you before you act on it.

Each of your questions appears in a bubble on the right, and each answer appears beside the assistant's icon. Point at one of your questions to see the time you sent it.

Under the message box, a line reads `The assistant can make mistakes — check important information. Actions always ask for your confirmation.` Click **Disclaimer** beside it to read what the assistant sees and where your data goes.

### Before you ask anything

An empty conversation greets you by your first name, for example `Hi Anna, how can I help?`, and says in one sentence what the assistant can help with. Under it are four questions you can start from: `Give me an overview`, `What approvals are waiting on me?`, `Request access to an application` and `Show my access review campaigns`. Click one to send it.

### Typing your question

Type your question in the message box at the bottom of the assistant.

* Press Enter to send it. Press Shift+Enter to start a new line instead.
* If you type with an input method, for example for Japanese or Chinese, the Enter that picks a word only picks the word. It does not send your question.
* The box starts two lines high and grows as you type, up to about two fifths of the screen. After that, it scrolls.
* A message has a maximum length, set by your organization. When you reach 80% of it, a counter such as `3200 / 4000` appears beside the send button. It turns red at the maximum, and the box takes no more text. If you paste a longer text, it is cut at the maximum.
* You can paste text, including cells you copied from a spreadsheet: their text goes into the box. The assistant does not take pictures or files, so pasting only a picture or a file puts nothing in the box.
* The send button, an arrow, is greyed out while the box is empty. While the assistant writes an answer, it becomes **Stop**: click it to stop the answer.

On a phone, the message box stays at the bottom of the screen, above the keyboard.

### Where your data goes

Click **More options** (the three dots at the top right of the assistant) to see the model the assistant runs on, and beneath it the line `Cloud model · data is sent outside your network`. It means that your questions, and the information the assistant looks up to answer them, are sent to a service outside your organization.

If no such line appears, the assistant has not said where it sends your data. Ask your IT administrator.

### Seeing how the assistant reached an answer

Before some answers, the assistant shows the reasoning it went through, folded away. Click it to read it.

To have that reasoning open every time, click **More options**, then turn on **Always show reasoning**. The choice is kept for this browser.

### Seeing what the assistant can do

To see what you can ask the assistant, and what it will not do, open **What can the assistant do?** in either of two ways:

* On an empty conversation, click the **What can the assistant do?** link under the suggested questions.
* Click **More options** (the three dots at the top right of the assistant), then **What can the assistant do?**.

A panel opens on the right and lists:

* **Answer** — what the assistant can tell you about: your approvals, requests and access; the roles, groups and applications you have; access review campaigns; your manager and the people who report to you; your sessions, devices and IT policy status; and how-to questions answered from the product documentation.
* **Act (always with your confirmation)** — what it can do for you once you confirm: approve or reject an approval task, and request access for yourself.
* **Open the right screen** — where it sends you instead of acting in the conversation: the pages for your password, multi-factor authentication (MFA) and security keys, and any other page for something it cannot do in the conversation.
* **Limits** — what it does not do: approve many items at once, change a password, MFA setting, security key or device, accept attachments or voice input, or see anything beyond what you are allowed to see.

The line about how-to questions appears only when your organization lets the assistant search the product documentation.

To close the panel, click **Close** (the cross at its top right), press **Esc**, or click anywhere outside it. On a phone, it fills the whole screen.

Until your organization's translation is available, the text in this panel is in English when you use SelfService in another language.

## Using the assistant with a keyboard or screen reader

You can use the assistant with the keyboard alone, and with a screen reader.

### What your screen reader tells you

Your screen reader names the conversation `Assistant`. It does not read an answer out word by word while the assistant is still writing it. Instead it tells you when something has happened:

* When an answer is complete, you hear `The assistant finished answering.` once. Move into the conversation to read the answer.
* After you approve, reject or request something from a card, you hear the short record that replaces the card, for example `Approved · 14:02. #1234 · Salesforce · Anna`.
* When the assistant cannot answer, you hear the title of the problem, for example `The answer took too long`. The problems are listed under [When the assistant cannot answer](#when-the-assistant-cannot-answer).
* After you copy something from an answer, you hear `Copied`.

### Where the cursor goes

Cards, suggested questions, sets of questions and notices come and go as the conversation moves on. If the cursor is on one of them when it disappears, the cursor moves to the message just above it, or to the conversation itself. It never jumps back to the top of the page, so you do not lose your place.

### Buttons that show only an icon

Buttons that show only an icon display their name when you point at them or move to them with Tab. These include **More options**, **Scroll to the latest message**, the send and stop button under the message box, and the buttons in your list of conversations, such as **New chat**. A screen reader reads the same name.

The buttons on code and tables in an answer, such as **Copy code**, **Copy table** and **Download table**, appear in the language you use SelfService in.

### Less motion

If you have turned on the setting that reduces motion on your device, for example **Reduce motion** on a Mac or iPhone, or turned off **Animation effects** in Windows, the assistant follows it:

* The shimmering line that says what the assistant is doing, such as `Thinking…`, shows as plain text.
* The conversation jumps to the latest message instead of scrolling to it smoothly.
* Panels, menus and other parts of the assistant appear and disappear at once, without animation.

The Segregation of Duties (SoD) warning on a confirmation card is also shown with enough contrast to read easily in the light theme.

## Following what the assistant is doing

While the assistant works on your question, a line above its answer says what it is doing, in plain words — for example `Thinking…`, `Looking up your approvals…`, `Checking your access`, `Searching organizations` or `Finding the right screen…`. If it takes longer than a few seconds, the line adds `Still working…`.

When the answer is ready, the line turns into a short summary with a check mark, for example `Looked up your approvals · Found the right screen`. Click the summary to see every step, in order.

A step that went wrong says so, for example `Could not load your approvals`, and the summary shows a warning instead of a check mark. When an administrator has switched off the assistant's access to your OpenIAM data, the step says that instead. A step that never finished, because you stopped the answer or it was cut off, shows `Stopped` in that list.

An answer that ended with a button to another SelfService page counts as finished. When you reload the page, or come back after following that button, the answer still shows its steps as done, and you can still regenerate it. Only an answer you stopped yourself shows `Stopped`.

Some of these lines may still appear in English when you use SelfService in another language, until your organization's translation is available.

## Asking about your access, roles and colleagues

You can ask the assistant about the access you have, about the roles, groups and resources in OpenIAM, and about the people you work with.

* **"Do I have the Finance role?"** The answer card is headed `You have this access` or `You do not have this access`. When you do not have it, the assistant also says whether you can request it, and if you can, the card offers **Request it**. Click it to request the access from the conversation; see [Requesting access for yourself](#requesting-access-for-yourself).
* **"What is the Finance role for?"** The assistant shows a card with the role's name, type, status and description. You can ask the same about a group or a resource.
* **"Find groups with Finance in the name."** You can search roles, groups, resources and organizations. Several results appear as a table; a single result appears as a card with its details.
* **"Who is my manager?"** The assistant shows your `Your profile` card, which names your manager next to `Manager`, and tells you who it is. Ask **"Show me my manager's details."** to see your manager's title and organization. If the assistant cannot look your manager up, it says that your manager is not available instead of guessing; you can still find your manager on your profile page, or ask your help desk.
* **"Show me Alex's details."** You can look up yourself, the people who report directly to you, and your own managers. When you search for people by name, the assistant finds only you and the people who report directly to you. It cannot find anybody else, such as a colleague, your manager's manager or a person who reports to somebody on your team, and it tells you so.
* **"What access does Sam Patel have?"** When you ask about the roles and groups of somebody who reports to you, the table's title names that person, for example `Access of Sam Patel (your direct report)`, so you can always see whose access you are looking at. If the name in your question is close to, but not, the name of somebody on your team, such as Sam Patil when your report is Sam Patel, the assistant usually asks whether you meant Sam Patel. Answer yes, or ask again with the name as it appears on your **My Team** page. Whatever name you typed, a table of somebody's access is always titled with that person's real name.

When you ask about somebody you are not allowed to see, the assistant tells you so in plain words instead of showing an error. If that person reports to you, it offers a button to your **My Team** page instead.

The assistant treats names, descriptions and other text stored in OpenIAM as information only. The same goes for the documentation pages it reads to answer a how-to question. If such text contains something that reads like an instruction, the assistant does not follow it.

## Reading the assistant's answers

Many answers come as a card or a table rather than only as text. The text that comes with them gives the total or a short conclusion, for example `You have 8 roles.`, and does not repeat what the table or card already shows: read the table for the items themselves.

* **Figures.** A question such as "How are my reviews going?" shows a row of tiles, each with one figure. Click a tile to open the page it comes from; the certification tiles open **My Campaigns**. Counts that are past due show in red, and counts that are due soon show in amber. If one figure could not be read, its tile shows `Could not load`, and the other tiles still show theirs.
* **Certification campaigns.** Each campaign shows as a card with its due date and the number of items you still have to review, for example `12 items left`, a progress bar such as `8 of 20 reviewed`, and an **Open campaign** button. When the assistant has room, the cards sit two to a row.
* **Your profile.** "Show my profile" shows a card with your initials, your name and status, your job title and organization, and your email address. Click **Open profile** to go to your profile page.
* **Pictures and diagrams.** The assistant never shows pictures. Where an answer, or its reasoning, contains one, you see only the picture's short description, and nothing is loaded from the internet. A diagram shows as the plain text it is written in.
* **Status labels.** Statuses such as `Pending`, `Approved`, `Rejected`, `Overdue` and `Active` show as colored labels. So does a person's status, for example `Pending Initial Login` or `Terminated`, on a person's details card and in a table of people, with the same words as elsewhere in SelfService. A status the assistant does not recognize shows as plain text, written as words rather than as a code: `Leave With Pay`, not `LEAVE_WITH_PAY`.
* **Links.** A link to a page in SelfService shows as the link's words followed by a button that opens that page, if you can use it. The button always carries the page's own name, for example **Pending approvals**, never words from the answer, so a button cannot claim to approve or change anything: it only opens the page. The button opens the page with only what that page needs, such as which task to show; anything else in the link, such as an address to send you to afterwards, is left out. A link to the OpenIAM documentation opens in a new tab. Any other link, for example to another website, shows as plain text that you cannot click. This keeps an answer from sending you somewhere unexpected. If you need that page, open it from the menu.
* **Pages the assistant sends you to.** When the assistant points you to another SelfService page, the answer shows a button with that page's name, for example **Pending approvals**. The button opens the page only when you click it: the assistant never opens a page for you. If you are not allowed to use that page, you see `This screen is not available to you` instead of the button.
* **Tables.** The first column shows an icon for the kind of item — a role or other access, a group, an application, a request or a task. A table of the people who report to you shows each person's initials. The row under your pointer is shaded, so you can follow it across. Long text, such as a long name or a reason, continues on the next line within its column, so you always see all of it. When a table is still wider than the conversation, for example on a phone, scroll it sideways. While there are more columns out of sight, the edge of the table fades out.
* **Examples from the documentation.** When you ask how something works, the answer can include an example from the product documentation, such as the text a system sends back, exactly as the documentation shows it.
* **Where an answer comes from.** An answer to a how-to question usually links the documentation pages it is based on, each shown as the page's title. Click one to read the page in a new tab. You can ask in your own language: the assistant searches the documentation, which is written in English, and answers you in your language.
* **Things only an administrator can do.** Some how-to answers describe work only an administrator can do, such as setting up a segregation-of-duties policy. The assistant tells you that your administrator does this and that you can ask them, rather than giving you the steps as your own.

While the assistant is still fetching the information, four grey lines show where the table or cards will appear.

### Seeing the details of a table row

When a table has more than you can read at a glance, such as a table of the requests waiting for your approval, you can open one row's details beside the conversation. This works when your browser window is wide: at least 1280 pixels, as on most computers with the window open fully. In a narrower window, the names in a table are plain text and there are no details to open.

1. Click the name in a row. With the keyboard, press Tab until you reach the name, then press Enter.
2. The row is tinted, and a **Details** panel opens on the right of the conversation.
3. Read the row's information in the panel. Each value sits under the same name as its column in the table. A status shows as the same colored label as in the table, and dates are written out in full.
4. To see another row, click its name. The panel switches to that row.

The panel shows only what is already in the answer. It does not look anything up again, so it is as up to date as the answer itself; to see the latest information, ask the assistant again.

The panel closes when you:

* click **Close details** at the top of the panel
* press Esc
* click **New chat**
* open another conversation
* make the browser window narrower

To open it again, click a name in the table.

## Picking a follow-up question

When the assistant has finished an answer, up to three suggested follow-up questions appear under it. While they are being prepared, pale placeholders pulse in their place. Click a suggestion to send it as your next question.

Suggestions only appear under the latest answer, and only when that answer finished normally — not under an answer you stopped, one that failed, or one the assistant could not put together. They also do not appear under an answer that ends on the record of an approval or rejection card, described in [After you decide](#after-you-decide). They disappear as soon as you send a new question, and they come back on the latest answer when you reopen the conversation. The questions on the empty assistant, before you ask anything, look the same and work the same way.

## Rating an answer or reporting a problem

Every finished answer has a row of small buttons under it, beside **Copy** and **Regenerate**. With a mouse or a trackpad, the row appears when you point at the answer or move to it with the Tab key. On a phone, a tablet or a computer with a touch screen, it is always shown. An answer with an approval, rejection or access request you confirmed has no **Regenerate**; see [After you approve, reject or request access](#after-you-approve-reject-or-request-access). Three of these buttons let you say how good the answer was:

* **Good answer** (thumbs up) — the answer was right and useful.
* **Bad answer** (thumbs down) — the answer was wrong or did not help.
* **Report a problem** (flag) — something in the answer needs a closer look, and you want to say what.

The buttons appear only once the assistant has finished the answer. While it is still writing, or while a card in the answer is waiting for you to confirm something, they are not shown yet.

### Rate an answer

Click **Good answer** or **Bad answer**. The button you chose stays highlighted, and `Thanks for the feedback.` appears.

To change your mind, click the other button. To take your rating back, click the highlighted button again; it is no longer highlighted, and the answer has no rating.

If you see `Your feedback could not be saved. Please try again.`, your rating was not kept, and the buttons go back to how they were. Click the button again a little later.

### Report a problem

{% stepper %}
{% step %}
Click **Report a problem** under the answer. The `Report a problem` dialog opens.
{% endstep %}

{% step %}
Optionally, describe what was wrong in **What went wrong? (optional)**. You can type up to 1,000 characters; the count under the box shows how many you have used.
{% endstep %}

{% step %}
Click **Send report**. Click **Cancel** instead to close the dialog without reporting anything.
{% endstep %}
{% endstepper %}

When the report is sent, `Thanks — the problem was reported.` appears, and the button now reads **Problem reported. Report it again**. You can report the same answer again later, for example to add something. Each new report replaces the comment you sent before.

If you see `The problem could not be reported. Please try again.`, the dialog stays open with your comment still in it. Click **Send report** again a little later.

Your comment is kept with this conversation. Do not put passwords or other secrets in it.

### What happens to your feedback

Your ratings and reports are saved with the conversation. They are still shown when you reload the page or come back to the conversation later, and you can rate answers in earlier conversations too. When the conversation is deleted, by you or automatically after the time your organization keeps conversations, your feedback is deleted with it.

Your organization can see which answers were rated and reported, to find out where the assistant needs to improve.

An answer that could not be saved to your conversation has no rating buttons; see [When an answer was not saved](#when-an-answer-was-not-saved).

## The language of the answers

The assistant answers in the language you write your question in, even when earlier questions in the same conversation were in another language. A spelling mistake does not change the language: "wat aproovals do i hav" gets an answer in English.

When a message does not show a language, for example only a name, a number or "ok", the assistant answers in the language you use SelfService in.

The suggested follow-up questions under an answer are in the same language as the answer. The names of your roles, groups, applications and requests stay as your organization stores them.

## How a conversation gets its name

A new conversation is first named after the first words of your question. Once the first answer has finished, the assistant gives it a short title of its own, in the language of that answer. If that first answer ends on the record of an approval or rejection card, the title comes with the next answer instead. If you renamed the conversation yourself in the meantime, your name is kept.

## Getting an overview

Ask the assistant for an overview, for example `Give me an overview` or `Summarize what I have and what is waiting on me`. It answers with one `Your overview` card that shows how many items wait for your approval, your review campaigns, the roles and groups you hold, and the applications you can open. Click a figure to open its page. While the assistant works, its activity line reads `Putting your overview together`.

If one figure cannot be found, that tile shows `Could not load` and the others still show. Ask again later for the missing figure, or open its page.

## Asking which approvals are overdue

Ask, for example, `Which of my approvals are overdue?`. The assistant counts the requests waiting for your approval whose due date has passed, and shows them in a table. A request without a due date is not counted as overdue.

When you have more requests waiting than the assistant can show at once, it does not say that all of them, or none of them, are overdue or come from one person based on the first few. It counts exactly what you asked about instead. To see every request waiting for you, open the **Pending approvals** page.

## Asking about the requests you made

Ask, for example, `What did I request last month?` or `Which of my requests are still open?`. For a period such as last month or this week, the assistant counts the requests you made in that period, open and finished alike, and shows them in a table.

Each row says what you asked for: the roles, groups or applications of the request, all of them, with a number such as `+3` when there are more than fit. A request that was not for access, such as a change to your profile, shows its description instead. The status shows whether the request is still in progress, completed, rejected or cancelled. Click a row to see the request in full. When there are more requests than the table shows, click **My requests** under it to see them all.

## Asking about the page you are on

The assistant knows which SelfService page you opened it from, and the language you use SelfService in. If you are looking at a task, a request, a certification campaign or a team member, it also knows which one. So on a request's page you can ask "What is the status of this?" without saying which request you mean.

It only knows which page and which item. It does not read what is on your screen, and the page is not kept with your conversation.

## Asking about your sign-ins, devices, accounts and IT policies

You can also ask the assistant about where you are signed in, the devices you registered, your accounts and the IT policies you have to accept. It always answers about you, never about somebody else.

* **"Where am I logged in?"** The assistant shows a `Where you are signed in` table with the application, the IP address and the time you signed in for each of your active sessions. A button takes you to the **My Sessions** page.
* **"Which devices do I have?"** The assistant shows a `Your registered devices` table with each device's manufacturer, model and platform. A button takes you to the **My Devices** page.
* **"What accounts do I have?"** The assistant shows a `Your accounts` table with each login you have on the systems your organization manages: the login, the system, whether the account is enabled, disabled or locked, and when its password expires. You can also ask "When does my password expire?". The assistant never shows a password or any other secret.
* **"Which policies do I still need to accept?"** The assistant shows a `Your IT policies` table. It lists first the IT policies you still have to accept for each application, how often you have to accept them again, and how long they have been waiting. Below them is the history of the policies you accepted or revoked, with dates. A button takes you to the **IT Policy** page, where you accept a policy yourself.

Statuses such as `Accepted`, `Revoked`, `Enabled`, `Disabled` and `Locked` show as colored labels. While the assistant looks these up, its activity line says so, for example `Looking up where you are signed in` or `Checking your IT policies`, and afterwards `Checked your IT policies`.

Some of these labels may still appear in English when you use SelfService in another language, until your organization's translation is available.

## Your list of conversations

On a wide screen, such as a computer, the assistant's page shows a **Conversations** panel on the left with all your conversations, and the conversation in the middle.

On a narrower screen, such as a tablet or a phone, the panel is hidden. Click **History** at the top left of the assistant to open it. It slides in from the left, and fills the whole screen on a phone. It closes when you open a conversation or start a new one. To close it without choosing, click **Hide conversations**, press Esc, or click beside it.

The list is grouped under `Pinned`, `Today`, `Yesterday`, `Last 7 days` and `Older`, by when each conversation last had a message. Pinned conversations always come first. The conversation you have open is highlighted. A conversation that has no title yet reads `New conversation`; when the assistant gives it a title during an answer, the title fades into the list. If your conversations do not all fit, click **Show more conversations** at the bottom of the list.

Click a conversation to open it.

### Finding a conversation

Type a word into **Search conversations** at the top of the panel. The list narrows to the conversations whose title or first message contains that word. If nothing matches, the list reads `No conversations found.`

### Finding an archived conversation

Under **Search conversations**, two buttons choose which conversations the list shows: **Active**, your list as usual, and **Archived**, the conversations you archived. Each time you open the assistant, or open the panel on a narrower screen, the list starts on **Active**.

Click **Archived** to see your archived conversations. The search works there too. If you have not archived anything, the list reads `No archived conversations.` Click an archived conversation to open it and read it, or to carry on with it.

To put an archived conversation back in your list, see **Unarchive** under [Renaming, pinning, archiving and deleting a conversation](#renaming-pinning-archiving-and-deleting-a-conversation).

### Renaming, pinning, archiving and deleting a conversation

Point at a conversation, then click **Conversation options** (the three dots beside it). On a phone, a tablet or a computer with a touch screen, the three dots are always shown beside every conversation. The menu offers:

* **Rename** — the title becomes a box you can type in. Press Enter to keep the new name, or Esc to leave the old one.
* **Pin** — moves the conversation to `Pinned`, at the top of the list. On a pinned conversation, the same item reads **Unpin**.
* **Archive** — removes the conversation from your list without deleting it. You find it again under **Archived**; see [Finding an archived conversation](#finding-an-archived-conversation).
* **Unarchive** — on an archived conversation, in place of **Archive**. The conversation leaves the archived list and is back in your list under **Active**.
* **Delete** — asks `Delete this conversation?` and names the conversation. Click **Delete** to remove it for good, or **Cancel** to keep it. If you delete the conversation you have open, the assistant starts a new, empty one.

### Deleting all your conversations

{% stepper %}
{% step %}
Click **History options** (the three dots at the top of the **Conversations** panel).
{% endstep %}

{% step %}
Click **Delete all conversations**.
{% endstep %}

{% step %}
The assistant asks `Delete all conversations?`. Type `DELETE` in the box. The **Delete all conversations** button in the dialog stays greyed out until you have typed the word exactly.
{% endstep %}

{% step %}
Click **Delete all conversations** in the dialog. Every conversation you have, archived ones included, is removed for good, and this cannot be undone. Click **Cancel** instead to keep them.
{% endstep %}
{% endstepper %}

### When a change is refused

If the assistant cannot rename, pin, archive, unarchive or delete a conversation, a notice above the list says why. Click **Dismiss** to close it.

* `This conversation is answering in another tab. Try again when it has finished.` — wait for the answer in the other tab to end, then try again.
* `This conversation no longer exists. It may have been deleted in another tab.` — the conversation is already gone.
* `The change could not be saved. Try again.` — try again in a moment.

If the panel shows `Your conversations could not be loaded right now.`, reload the page.

### Hiding the panel

On a wide screen, click **Hide conversations**, the first button at the top left of the assistant, to hide the panel and give the conversation more room. Click the same button, now **Show conversations**, to bring it back. The browser remembers your choice the next time you open the assistant.

## Coming back to a conversation

The quickest way back to a conversation is to click it in the **Conversations** panel. You can also use its address: as soon as the assistant gives its first answer in a new conversation, the address in your browser changes to one that belongs to that conversation. Opening that address later brings the whole conversation back.

{% stepper %}
{% step %}
While the conversation is open, bookmark the page, or copy its address.
{% endstep %}

{% step %}
When you want to return to it, open the bookmark or paste the address into your browser.
{% endstep %}

{% step %}
Type your next question at the bottom and carry on.
{% endstep %}
{% endstepper %}

The address only opens the conversation for you. If you send it to a colleague, they see that the conversation could not be found, because nobody else can read your conversations.

If you leave the assistant and come back to it in the same browser tab, it reopens the conversation you last had open in that tab. A new tab, or a tab you closed and opened again, starts with an empty conversation.

## The same conversation in two tabs

If you have the same conversation open in two browser tabs, the assistant answers in only one of them at a time. While one tab is answering, the other shows `This conversation is answering in another tab.` above the message box, and its message box, **Send** and **Regenerate** are greyed out.

When the answer ends, the waiting tab shows the new messages by itself, without a reload, and you can type in it again. If the answering tab was closed or stopped working in the middle of an answer, the other tab becomes usable again within about a minute.

Tabs that show different conversations do not affect each other.

## Starting a new conversation

Click **New chat** at the top of the assistant, beside its title, or click **+** at the top of the **Conversations** panel. On a phone, **New chat** shows only its icon. The assistant opens an empty conversation, and the one you left stays saved; you can still return to it from the list or from its address.

## Reading earlier messages

A long conversation opens on its latest messages. Scroll up to load the ones before them, a page at a time. You can also click **Show earlier messages** at the top of the conversation.

If you see `Earlier messages could not be loaded.`, click **Show earlier messages** again.

If you start typing a question and then switch to another conversation, your unsent text is still there when you come back, as long as you do not reload the page.

## Cards from an earlier visit

Some answers come with buttons, for example a card asking you to confirm that you approve a request, or a short set of questions the assistant needs answered. Those buttons only work while the card is new and is the last thing in the conversation.

When you reopen a conversation, or when the conversation has moved on past a card, a card that was still waiting for you shows `No longer active — ask again` instead of its buttons. Nothing was approved, rejected or sent from it. To do what the card offered, ask the assistant again, and it gives you a fresh card.

A card you cancelled keeps its record, `Cancelled` and the time, when you reopen the conversation. That includes a card you cancelled by sending a new message while it waited.

The exception is a card you had already confirmed when you left or reloaded the page, while your decision or request was still being saved. That card shows `Checking whether this went through…` instead, and then what really happened. See [When a decision or request is interrupted](#when-a-decision-or-request-is-interrupted).

The buttons on a set of questions from the assistant read **Submit** and **Cancel**, in the language you use SelfService in.

## Approving or rejecting a request

When a request waits for your approval, you can decide it from the assistant. Ask, for example, "Approve request #1234" or "Reject the Finance-Read request". The assistant never decides a request on its own. It shows a card for you to check first, and nothing happens until you press a button on that card.

### What the card shows

Everything on the card comes from the request itself, not from the assistant's own words:

* **What will happen.** A title and one sentence saying what your decision does, worded for the kind of request:
  * A request for access is titled `Approve access request`, and the sentence reads, for example, `Anna Kovac will get Finance-Read in SAP until 31 Dec 2026.` If you reject it, it reads `Anna Kovac will not get Finance-Read in SAP.`
  * A request to remove access is titled `Approve access removal`, and the sentence reads, for example, `Anna Kovac will lose Finance-Read in SAP.` If you reject it, it reads `Anna Kovac will keep Finance-Read in SAP.`
* **The details.** **Requester**, **Item**, **Beneficiary**, **Justification** and, for a request for access, **Duration**. A request with no end date shows `No end date`.
* **Validation.** Either `No policy conflicts found.`, or a `SoD conflict` warning that names the Segregation of Duties (SoD) rule the request breaks, for example two kinds of access that one person should not hold together. When the request was not checked, the card says `Policy conflicts were not checked for this request.`
* **Open approval task.** A button, for example **Open approval task #1234**, that opens the request's task page, where you can see everything about it.

While the card reads the request, it shows `Loading the request…`. If it shows `Could not load this request.`, click **Try again**.

### Approve the request

{% stepper %}
{% step %}
Read the sentence and the details on the card.
{% endstep %}

{% step %}
Click **Approve request** followed by the request's number, for example **Approve request #1234**.
{% endstep %}
{% endstepper %}

### Reject the request

{% stepper %}
{% step %}
Click **Reject**.
{% endstep %}

{% step %}
In **Reason for rejection**, type why you are rejecting it. The requester will see this comment, and the reason is required.
{% endstep %}

{% step %}
Click **Reject request** followed by the request's number, for example **Reject request #1234**.
{% endstep %}
{% endstepper %}

If you asked the assistant to reject the request, the card opens with **Reason for rejection** already showing. If you click **Reject request** with no reason, the card shows `Please add a reason.` and sends nothing.

### Change your mind

Click **Cancel**, or press Escape while the card is selected. Nothing is approved or rejected.

While **Reason for rejection** is showing, **Cancel** only takes you back to the first set of buttons, so you can approve instead. Press Escape to cancel the whole card.

### Using the keyboard

When the card appears, the card itself is selected, never one of its buttons. So if you are still typing and press Enter, nothing is approved or rejected by accident. Press Tab to move to the buttons, then Enter to press one.

### After you decide

The card is replaced by a short record of what you did, for example `Approved · 14:02`, `Rejected · 14:02` or `Cancelled · 14:02`. Below it are the request's number, the item and the beneficiary, the reason you gave for a rejection, and `Request status:` followed by the request's status as it is now. Click **Open** to go to the request's task page. A cancelled card has no **Open** button.

The record is the assistant's whole statement of the outcome: the answer ends there, and the assistant does not add a sentence repeating what the record says.

After you decide, the cursor goes back to the message box, so you can type your next question straight away.

If your decision could not be saved, the record reads `Could not complete`, followed by `The server returned:` and the reasons OpenIAM gave. Nothing changed. Open the task page from the record to decide the request there, or ask your help desk if the reason is not clear. If OpenIAM gave no reason, the assistant adds a short explanation under the record.

If the connection was lost before OpenIAM answered, the record can also read `Outcome unknown`. See [When a decision or request is interrupted](#when-a-decision-or-request-is-interrupted).

### When the request cannot be decided here

The assistant decides only a request that gives one person one item, or removes one item from one person, because only then can the card say exactly what your decision does. Any other request is decided on its task page, where you see all of it.

When a request cannot be decided from the assistant, the card says why instead of showing the approve and reject buttons, and offers **Open approval task** to decide it on its task page:

* `This request is no longer waiting for your decision.` — somebody already decided it, or it was closed.
* `This request lists several items. Open its task page to review and decide each of them.` — the request asks for more than one item, so you review every item on its task page.
* `Only a request to grant or remove one item for one person can be decided here. Open its task page to decide this one.` — any other kind of request, for example a change to a group or a role, a new user, or a request that does not name the person it is for.
* `Claim this request on its task page before you decide it.` — the request has to be claimed on its task page first; once you have claimed it there, you can decide it.
* `This request cannot be decided here. Open its task page instead.` — for any other reason.

A card for a request it cannot decide is titled `Approve request` or `Reject request`, and it has no sentence saying what a decision would do.

### From a table of your approvals

When you ask what is waiting for your approval, the assistant shows a table of those requests: up to the first 50, also when you only ask how many are waiting, and the answer gives the number. When you ask which of them are overdue, the assistant counts the requests whose due date has passed; a request with no due date is never counted as overdue. Each row has **Approve** and **Reject** buttons. They do not decide the request straight away: they ask the assistant for you, for example "Approve request #1234", and the assistant then shows the card described above for you to check. The buttons only work in the latest answer.

## Requesting access for yourself

You can ask the assistant for access you do not have yet. It fills in the same request you would make on the [request screen](https://docs.openiam.com/end-user-guide-for-selfservice/requests/request-access.md), and it never sends one without your say-so: nothing is requested until you press **Submit request**.

### Finding what you can request

Ask, for example, "What can I request for Salesforce?" or "What can I request?". The assistant shows a `What you can request` table of the applications you can request, with their **Name** and **Category**. Each row has a **Request** button, which asks the assistant for you, for example "Request Salesforce for myself".

The table can also show three more columns, when your organization has recorded these details for the applications:

* **Type** — what kind of application it is.
* **Risk** — how sensitive the access is, as a colored `High`, `Medium` or `Low` label.
* **Owner** — the person or group responsible for the application.

A column appears only when at least one application in the table has a value for it. An application with no value leaves its cell empty.

### Request access

{% stepper %}
{% step %}
Ask for the access, for example "Request Salesforce for myself". You can also click **Request** on a row of the table, or **Request it** on an answer that says you do not have some access.
{% endstep %}

{% step %}
A form headed `Request` and the item's name appears in the conversation. If the application offers more than one kind of access, choose the one you need under **Which access do you need?**.
{% endstep %}

{% step %}
In **Justification**, type why you need this access. The justification is required.
{% endstep %}

{% step %}
Under **Duration (optional)**, choose how long you need the access. The form only offers the lengths this access allows: **No end date**, **30 days** and **90 days**, or, when the access has a time limit, the lengths within it and a **Maximum** option. When the access cannot be time-limited, the form has no duration.
{% endstep %}

{% step %}
Click **Continue**. The form turns into a summary of what you entered, marked `Details sent`.
{% endstep %}

{% step %}
Wait while the assistant shows `Checking policy rules`. It then reads `No conflicts`, or the number of conflicts it found, for example `1 conflict found`.
{% endstep %}

{% step %}
Read the confirmation card, then click **Submit request**.
{% endstep %}
{% endstepper %}

### What the confirmation card shows

* **What will happen.** The title `Request access`, and one sentence that says who will approve the request first, such as `A request for Sales-User in Salesforce will be sent to Jana Novak for approval.` Instead of a person, the approver can read `members of the Finance Approvers role`, `members of the Finance Approvers group` or `your manager`. When several people share that first approval, the sentence adds how many, for example `Jana Novak or one of 2 other approvers`. When OpenIAM cannot tell who approves, the sentence reads `A request for Sales-User in Salesforce will be sent for approval.`
* **The details.** **Item**, **Beneficiary** (always `You`), **Justification** and **Duration**.
* **Validation.** The same notice about Segregation of Duties (SoD) conflicts that the request screen shows. When the access conflicts with a policy but can still be requested, tick the box saying you understand the conflict. **Submit request** stays greyed out until you do.
* **Open the request screen.** A button that takes you to the request screen, if you would rather make the request there.

To change your mind, click **Cancel**, or press Escape while the card is selected. Nothing is requested.

### After you submit

The card is replaced by a short record of what happened:

* The request's number and the time, for example `Request #4711 submitted · 14:02`, with the item and the duration below. Click **Open request** to follow the request. Your approvers take it from there, as for any other request. When OpenIAM does not return the request's number, the record reads `Request submitted` and offers **Open my requests** instead.
* `Could not complete`, followed by `The server returned:` and the reasons OpenIAM gave. No request was sent. Click **Open the request screen** to request the access there, or ask your help desk if the reason is not clear. The request screen opens with that application already selected and its list of access open, so you can pick the access and try again. If the application cannot be opened for you there, the request screen opens as usual with the note `This item cannot be opened here`. Click **Dismiss** to close the note, then find the access in the catalog.
* `Cancelled` and the time, with `No changes were made.`
* `Outcome unknown` and the time, when the connection was lost before OpenIAM answered and the assistant could not find your new request. See [When a decision or request is interrupted](#when-a-decision-or-request-is-interrupted).

If the record adds `The item is still in your request cart. Remove it there so it is not requested again.`, open the request screen and remove the item from your cart.

### When the item cannot be requested here

Some access can only be requested on the request screen. Instead of a form, the assistant then says why and offers **Open the request screen**. For example:

* `You already have this access.`
* `This access is already in your request cart. Finish the request on the request screen.`
* `This access needs choices that only the request screen collects.`
* `This item can only be requested on the request screen.`
* `This item is not in your request catalog. Look for it on the request screen.`
* `This item offers more access than fits here. Choose it on the request screen.`

### Requesting access for somebody else

The assistant only requests access for you. If you ask it to request access for somebody else, for example "Request Salesforce for Anna", it creates nothing. It offers a **Request for others** button instead.

Nothing opens until you click the button. The assistant offers it straight away, without looking the person up first, so you get the button even for somebody who does not report to you. Usually the button opens **Create Request**; select **Change** there to pick the person. When the conversation has already shown who you mean, for example after you looked at your team, the button can open the request screen with that person's name next to "Requesting for" instead, so all that is left is to choose what they need. See [Requesting for other people](https://docs.openiam.com/end-user-guide-for-selfservice/requests/request-for-others.md).

You see the button only if you are allowed to request for other people. Otherwise the assistant shows `This screen is not available to you`.

To add one of your direct reports to a group or role, the assistant may offer **My team** instead of **Request for others**. It opens your **View Direct Reports** page. Two buttons at the top right of that page act on several people at once. Each appears only if your organization allows it for you:

* **Bulk Request** — tick the people, select **Continue**, and choose the group or role on the **Create Request** page that opens. You can tick only active people who report directly to you. Like any request, it still needs approval. See [Requesting for several people at once](https://docs.openiam.com/end-user-guide-for-selfservice/requests/request-for-others.md#requesting-for-several-people-at-once).
* **Bulk Operations** — for contractors only: you can tick only active contractors. Tick them and select **Continue**. Under **Operation**, choose **Add Entitlement**. Under **Entitlement Types**, choose **Group** or **Role**, search for the one you want, and select **Submit**.

If you cannot open **View Direct Reports**, the assistant shows `This screen is not available to you` instead of **My team**.

## Answering the assistant's questions

Sometimes the assistant needs more from you before it can help, for example which of two roles you mean. It then shows a short set of questions under the heading `The assistant asks`, each with choices to pick from. The set is selected when it appears, never one of its buttons, so a stray Enter does not send it. Press Tab to move to the first choice.

The questions and choices are the assistant's own words. The heading and the two buttons, **Submit** and **Cancel**, are always the same, in the language you use SelfService in.

The set has no box to type in: you only pick from its choices. When the assistant needs an answer in your own words, such as a reason, it asks in its reply instead, and you answer in the message box as usual. A set of questions that asks for a password, PIN or one-time code is never shown. If the assistant's reply ever asks for one, do not type it: what you send is passed to the assistant and kept in the conversation. Tell your help desk instead. To change a password or set up a sign-in method, use the pages for that in SelfService.

If you click **Submit** with a question unanswered, nothing is sent: that question shows `Answer this question to continue.`, and the cursor moves to it.

To answer with the keyboard:

* Each list of choices is a single Tab stop. Press Tab to move into the list, and Tab again to move on to the next question.
* Use the arrow keys to move between the choices in a list. Moving to a choice does not pick it.
* Press Space or Enter to pick the choice you are on.
* Press Esc to go back to the message box without sending. The answers you have already given stay in the set, so you can come back and finish it.

## One decision at a time

The assistant never approves, rejects or requests anything on its own: it always shows a card for you to confirm first. It offers one such card per answer. While a card waits for you, and after you have confirmed or cancelled it, the assistant does not offer another card in the same answer.

So if you want to approve two requests, confirm the first card, then ask the assistant about the second one in a new message. If you ask it to approve several at once, it explains that each approval is confirmed on its own card instead. It only decides tasks that are in your approvals, so it does not act on a task number you type that is not one of them.

If you cancel an approval or rejection card, its record reads `Cancelled` and the answer ends there. Nothing is approved or rejected.

If you paste a one-time code into the conversation for an approval, the assistant refuses it. It offers the page where you set up your authenticator only when you ask to set up or reset a sign-in method.

The assistant also only ever looks up and acts on your own requests, approvals and access. You cannot ask it to act as somebody else. If you manage a team, you can still ask about the roles and groups of the people who report to you. When you ask about anybody else, for example a colleague's requests, your manager's roles or a user ID you type, the assistant tells you that it can only show your own information and your team's roles and groups. It does not show your own information in its place, and it does not offer you a button to one of your own pages instead.

## Sending a message while a card is open

You can type a new message while a card still waits for you, or while a decision you confirmed is still being saved. What happens to the card depends on whether you had already confirmed it.

* **You had not confirmed the card yet.** Sending a new message cancels the card. It shows `Cancelled` and the time, and nothing is approved, rejected or requested. The assistant then answers your new message, and it knows that you cancelled the card. The card and your earlier question stay in the conversation, above your new message: scroll up to see them.
* **You had already confirmed the card.** Your approval, rejection or access request is not cancelled. Your new message waits until it has been saved, and the card then shows what really happened: the request approved or rejected, the access request you submitted with its number, or `Could not complete` with the reasons OpenIAM gave. For an access request, the assistant first tells you that outcome, sometimes in a short extra answer; for an approval or rejection, the record on the card is the outcome. The assistant then answers your new message.

While your first message waits, a second one is not sent. Its text goes back into the message box, so you can send it once the assistant has answered. The same happens if the assistant cannot pass on the outcome of your decision: your message is not sent and its text stays in the message box. Send it again.

## When you are asked to verify your identity

Your organization can ask you to verify your identity before an approval, rejection or access request goes through. The card then waits while you verify, and your decision or request is sent as soon as you have. If you close the verification window instead, nothing is approved, rejected or requested: the card shows `Not submitted: identity check cancelled.` and keeps what you entered, so you can confirm it again or cancel it. If the card shows `Verification didn't take effect. Try again.`, confirm it again.

## When a decision or request is interrupted

Once you confirm a card, your approval, rejection or access request goes to OpenIAM. Sometimes the answer from OpenIAM never arrives, for example because your connection drops. Your decision or request may still have been saved, so the assistant does not report it as failed. It looks at the request again and shows what it finds.

### For an approval or rejection

* **Your decision was saved.** The record reads `Approved` or `Rejected` and the time, as usual.
* **The request still waits for you.** The record reads `Could not complete`, with `The connection was lost before the server answered. The task is still waiting for your decision, so nothing was recorded.` Nothing changed, and you can decide the request again.
* **The assistant cannot tell.** The record reads `Outcome unknown` and the time, with `The connection was lost before the server answered, so the decision may or may not have been recorded. Check the task before you decide again.` Click **Open approval task**, for example **Open approval task #1234**, to see on the task page whether your decision was saved.

### For an access request

* **Your request was saved.** The record reads `Request #4711 submitted` and the time, as usual.
* **The assistant cannot find it.** The record reads `Outcome unknown` and the time, with `The connection was lost before the server answered, so the request may or may not have been submitted. Check your requests before you ask again.` Click **Open my requests** to look for it. Ask for the access again only if your request is not there and you do not have the access yet.

With `Outcome unknown`, the assistant does not report your decision or request as failed.

### Leaving or reloading the page

While your decision or request is being saved, your browser asks you to confirm before you leave or reload the page. It is safest to stay until the card shows its record.

If you reload the page anyway, the card shows `Checking whether this went through…` when the conversation opens again. The assistant looks at the request again, and the card then shows one of the records above. A card that you had not confirmed is not affected: it shows `No longer active — ask again`, and nothing was approved, rejected or requested from it.

## When an answer was not saved

Occasionally an answer appears in full, but the assistant cannot save it to your conversation. It then shows `This answer could not be saved to your conversation history.` under the answer.

The answer itself is still correct and you can use it. Only the record of it is missing, so it would not be there the next time you open the conversation. You cannot rate an answer that was not saved, or report a problem with it.

{% stepper %}
{% step %}
Click **Retry** beside the note. **Retry** appears only on the latest answer, once the assistant has stopped writing.
{% endstep %}

{% step %}
Wait for the note to change. `Saved to your conversation history.` means the answer is now kept. `Still not saved. Try again in a moment.` means it is not, yet; click **Retry** again a little later.
{% endstep %}
{% endstepper %}

## When a conversation will not open

* `This conversation could not be found. It may have been deleted.` — the conversation no longer exists, or the address belongs to somebody else's conversation. Your organization deletes old conversations automatically after a set time. Click **New chat** to start again.
* `This conversation could not be loaded right now.` — the conversation exists, but could not be read at the moment. Click **Retry**.

## When the assistant cannot answer

If something goes wrong while the assistant works on your question, it shows a shaded box with an icon, a title that says what kind of problem it was, and a sentence explaining what happened:

* `The assistant is unavailable` — the assistant could not be reached.
* `An OpenIAM lookup failed` — the assistant could not read some of your information, for example your pending approvals. The assistant sometimes tries a lookup a second time on its own. If the second try works, you see the answer and no box; the box appears only when the information still could not be read, once the assistant has finished answering.
* `Too many requests` — the assistant, or the service it runs on, is handling too many questions right now.
* `Your session expired` — you have been signed out.
* `The answer took too long` — the answer did not arrive in time.
* `The assistant ran into a problem` — anything else.

If the assistant finished without showing you anything — no text, card, table or button — the box reads `The assistant could not put together an answer. Please try again.`

If your message is longer than your organization allows, the assistant shows `Your message is too long. Shorten it and send it again.` Your text stays in the message box, so you can shorten it and send it again.

What you can do depends on the problem:

* Click **Retry** to ask the same question again.
* When the box says `Too many requests`, **Retry** stays greyed out and reads `Try again in` and a number of seconds, counting down. Wait until it becomes available, then click it.
* When your session expired, click **Sign in again**.
* When the assistant could not read your information, the box may also offer a button such as **Open Pending approvals**, which takes you straight to that page so you can look for yourself. It appears only if that page is in your SelfService menu.
* When the box says `An administrator has switched off the assistant's access to OpenIAM data, so it cannot look this up right now. You can still open the screen that shows it.`, it offers no **Retry**, because asking again does not help until the access is switched back on. Use the button to that page instead, when the box offers one. The assistant still answers questions that do not need your OpenIAM data, and about a minute after the access is switched back on, it can look your information up again.
* When the problem comes right after you confirmed a card, the box offers **Continue** instead of **Retry**. See [After you approve, reject or request access](#after-you-approve-reject-or-request-access).

### After you approve, reject or request access

Once you confirm a card, your approval, rejection or access request has gone to OpenIAM, and the card's record shows what happened. Asking the same question again would take that record out of the conversation, and the assistant could show you a new card for the same request. So an answer with a confirmed card never offers that:

* The answer has no **Regenerate** button, from the moment you confirm the card.
* When something goes wrong after you confirm the card, the box offers **Continue** instead of **Retry**. Click it to let the assistant finish its answer. The record stays as it is, and nothing is approved, rejected or requested again.
* When you stop the answer, the `Stopped` line under it offers **Continue** instead of **Regenerate**.

**Continue** appears once the card shows its record. After `Too many requests`, it counts down like **Retry** does.

If the same conversation is open in another tab that still shows the card as it was before you decided, and you click **Regenerate** there, the assistant refuses with `This answer includes a decision or access request you made, so it cannot be regenerated. Open the conversation again to see it as it was saved.` The full assistant page then opens the conversation again by itself, with the record of your decision or request.

## Need help?

If a conversation still will not open, or answers keep showing that they could not be saved, contact your help desk or IT administrator. Tell them roughly when it happened and what you were asking about.


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://docs.openiam.com/end-user-guide-for-selfservice/ai-assistant-conversations.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
