> ## Documentation Index > Fetch the complete documentation index at: https://learn.nextedy.com/llms.txt > Use this file to discover all available pages before exploring further. # Admin AI Assistant > How the Admin AI Assistant works in Nextedy RISKSHEET: a documentation-grounded chat in the configuration editor, scoped to the Risksheet documentation, and designed so the editor keeps working even when the assistant does not. export const LastReviewed = ({date}) => { if (!date) return null; const formatted = new Date(`${date}T00:00:00Z`).toLocaleDateString("en-US", { year: "numeric", month: "long", day: "numeric", timeZone: "UTC" }); return

Last reviewed on {formatted}

; }; The **Admin AI Assistant** is a documentation-grounded chat in the Risksheet configuration editor. It is not included with Risksheet: it arrives with the separately installed **Nextedy AI** package. It opens from the sparkle button in the editor toolbar, and its window is titled **✦ Risksheet AI**. While an administrator is editing a Risksheet configuration, they can ask natural-language questions about Risksheet configuration and get answers drawn from the Risksheet documentation, with links to the relevant pages. This page explains what the assistant is, why it works the way it does, how its answers stay grounded in the documentation, and what determines whether it is available. For the task of opening and using it, see [Use the Admin AI Assistant](/risksheet/guides/configuration/use-admin-ai-assistant). **One assistant, shared across products** The assistant is not part of Risksheet itself. It arrives with the separately installed **Nextedy AI** package, which plugs into the configuration editor that Nextedy products share. The Risksheet build tells the assistant which product it is serving, so the panel is titled **✦ Risksheet AI** and answers are drawn from the Risksheet documentation rather than another product's. **Add data model**, offered in the Powersheet editor, is not available here — Risksheet has no separate data-model file to attach. ## Why it exists Switching context to a separate documentation site or a support channel breaks the configuration workflow and slows down problem solving. The assistant puts the answer one click away from where the question arises — inside the editor, while the configuration is still on screen — so administrators spend less time hunting through documentation and more time configuring. The questions it is built for are the ones that come up mid-task: how to configure a particular behavior, which property controls it, what an existing block of configuration does, why something is not working, the difference between two concepts, or simply where something is documented. ## A guide, not an operator The assistant answers questions and points to the right documentation. It does **not** change, save, or apply configuration on your behalf. The administrator stays in control of the configuration file and makes every edit themselves in the editor. There is no code path from the chat to your saved configuration. **Two assistants, both labeled AI Assistant** Nextedy AI adds two assistants to Risksheet, and the interface labels both of them **AI Assistant**. This page describes the **Admin AI Assistant** — the one in the configuration editor, which only reads documentation and answers questions about configuring the product. Its panel is titled **✦ Risksheet AI**, and it never changes anything. The other is the **Risksheet AI Assistant**, opened from a sparkle button in the Risksheet grid toolbar, which works on the risk content itself. Both reach your installation through the same Nextedy AI plugin, but they take it from there differently — the Admin AI Assistant asks the plugin only for the proof it needs to reach the Nextedy AI documentation service, while the Risksheet AI Assistant runs its actions against the Polarion server. ## Grounded in the documentation The assistant runs in a documentation-search mode rather than as an open-ended chatbot: its answers are grounded in documentation content, not generated freely. Several design choices keep the answers relevant: | Aspect | How it works | | --------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Product scoping** | Every request identifies the product as Risksheet, so the service answers for Risksheet rather than for another Nextedy product. | | **Streaming responses** | Answers stream in incrementally as they are generated. A typing indicator is shown between sending the question and the first content arriving. | | **Conversation continuity** | The conversation is held on the Nextedy AI service, so a short follow-up is understood against the earlier turns, and the conversation is restored when you reopen the panel. | | **Starter questions** | When the panel is empty it offers suggested starter questions, fetched from the AI service for the current product. If the service returns none, no chips are shown. | Conversations do not live forever. The editor keeps only the conversation's identifier in the browser; the turns themselves are held by the AI service, and once it has dropped a conversation the panel simply starts a fresh one. Because answers come from a language model reading documentation, the assistant shows a standing reminder that *"AI can make mistakes. Verify important information."* Treat its answers as guidance and confirm property names and behavior against the linked documentation before relying on them in production. ## What the assistant sees By default the assistant answers from the documentation alone — it does **not** receive your configuration. When a question is specifically about the file being edited, the administrator can explicitly attach it with **Add configuration**, and only then is that content sent alongside the question. The attachment appears as a **Configuration** chip and can be removed at any time. **Privacy by default** Sending no configuration content unless it is explicitly attached is deliberate: attaching your configuration is a decision you make per conversation, not something the assistant does for you. On the default path your configuration stays out of the request entirely, and the assistant reads only what it needs to search the documentation. ## Availability The editor resolves an availability status before enabling the assistant, and reflects it on the toolbar button. The signal is deliberately **not** a product health check — it is the result of the token flow described below actually succeeding. | State | What it means | In the editor | | ------------------------- | ----------------------------------------------------------------------------------------------------------- | ---------------------------------------------------------------------------------------------------------------------------- | | **Available** | The Nextedy AI package answered and a session token was issued. | The **AI Assistant** button is enabled; its tooltip reads *"AI Assistant"*. | | **Package not installed** | The Nextedy AI package is absent from the server. | The button is disabled; its tooltip reads *"Install the Nextedy AI plugin, or contact Nextedy, to enable the AI Assistant"*. | | **Unavailable** | The package is installed but could not be reached, or the proof could not be exchanged for a session token. | The button is disabled; its tooltip reads *"AI Assistant is currently unavailable"*. | ### How the browser authenticates Authentication runs in two hops, both made by the **browser** — the Polarion server itself never makes an outbound call: The editor calls the Nextedy AI plugin on your own Polarion server at `/polarion/nx_ai/api/ai-token`. Polarion returns a signed license proof for the signed-in user. This request stays inside your network. The browser posts that proof to the Nextedy token service, which validates it and returns a short-lived session token. Every question is sent to the Nextedy AI service with that token attached. The token is cached in the browser and refreshed automatically shortly before it expires. Because the proof travels through the browser, the assistant works even when your Polarion server has no outbound internet access. **No licence is needed, and the button never claims otherwise** The Admin AI Assistant requires **no** valid Nextedy AI license. Installing the Nextedy AI package on the server is enough, together with a signed-in Polarion user — which is why none of the three tooltips above mentions a licence. **This applies to the editor assistant only.** The Risksheet AI Assistant in the sheet does require a valid license: on the grid toolbar an invalid one disables the button with its own, differently worded tooltip. See [Nextedy AI](/overview/resources/nextedy-ai) for what the license covers. **The button appears even without the package — and can be turned off** When the Nextedy AI package is not installed, the **AI Assistant** button still appears in the editor toolbar, disabled, rather than being hidden, and its tooltip reads *"Install the Nextedy AI plugin, or contact Nextedy, to enable the AI Assistant"*. This is deliberate: it advertises the capability and says how to obtain it. An administrator who does not want it advertised can turn it off through context properties, independently for the editor assistant and for the one in the sheet. ### Graceful degradation Availability of the assistant is an independent signal from the rest of the editor. If the Nextedy AI plugin or service cannot be reached, only the **AI Assistant** button is disabled — the configuration editor, and the rest of Risksheet, continue to work normally. The assistant is an addition to the editor, never a dependency of it. ## Related Pages Step by step: open the assistant in the editor, ask questions, and attach the configuration you are editing. Open the editor the assistant lives in, and save configuration changes. The package the assistant arrives with, where it appears in each product, and what its license covers. How templates and document configurations inherit and override each other. The property reference the assistant's answers draw on.