Webinar
How to create your own AI assistants directly in Sitevision
AI that understands your content. That provides quick answers. And makes everyday life easier – for both employees and visitors.
Speakers
In this free webinar, Alexander Gustafsson shows you how to create your own AI assistants directly in Sitevision. The assistant is trained using your information and quickly becomes a helping hand for anyone who needs answers, whether they work for you or are searching for something on the web.
You will see how AI assistants can:
- Understand your language and provide quick answers.
- Take the load off by handling recurring questions.
- Be trained on your own content and controlled in both knowledge and behavior.
- Be used on both intranets and websites.
- Raise the level of service for both employees and visitors.
We conclude with an open Q&A session where you can get answers to your specific questions.
Frequently asked questions about Sitevision AI
We received many questions during the webinar. We think that's great 🎉! Here we have tried to answer them all.
How much does the assistant cost?
For pricing information, please refer to the form at www.sitevision.se/assistent. There you can easily submit a request and receive information about the cost.
Does it work for those of us who have on-prem (own operation)?
Yes, the Assistant works both for those who run Sitevision on their own servers (on-prem) and for those who use Sitevision Cloud.
Is the idea that the assistant only works on intranets and not external websites?
The assistant can be used both on intranets and external websites.
What is required in order to use a source other than Sitevision or documents in Sitevision's file archive?
There are APIs for feeding the knowledge base with data from sources other than Sitevision and documents in Sitevision's file archive. In addition, we plan to offer ready-made integrations via Sitevision Marketplace, which makes it easy to connect to external systems.
What is the crawling/indexing frequency for the assistant?
Pages, articles, and documents included in a knowledge base are updated immediately upon publication. This means that the Assistant always has access to the latest version of the content without delay.
Could we use, for example, AI Sweden's Svea solution as an AI service?
Yes, if you need to use your own AI models, such as AI Sweden's Svea solution, it is possible to connect these. Please contact us for more information about using your own language models.
Is it possible to configure it to always use the most recent information so that it does not mix it with old information in its response?
The assistant always prioritizes newer information in its responses. However, it is a good idea to keep track of the content and, as far as possible, ensure that outdated information is not retained in the assistant's knowledge base. This minimizes the risk of old information being mixed in.
Which AI model is used and where is it operated?
We use our own language models (LLMs) in Sweden. This means that we do not send data to external services such as OpenAI, Azure, Google, or similar. We continuously update with new models and versions, depending on what provides the best performance and security for our customers.
We have Sitevision AI but are missing Assistant and Knowledge Base under website settings. Is this something that can be added free of charge by contacting support?
The assistant is an add-on and is not automatically included. We are making significant investments in brand new infrastructure to be able to offer an AI service that is secure, Swedish, and fully compliant. With our solution, no data is sent to third countries, and it is not used for general model training, all to ensure the highest possible integrity and security.
Unlike many other AI services, which are priced based on so-called tokens and make it difficult to predict costs, we can offer AI assistants at a fixed monthly cost. This gives you full control and makes it easier to plan your budget. We want you to be able to use AI in a smooth and predictable way, without unnecessary surprises.
For an estimate of the cost of the add-on, please contact your account manager at Sitevision.
As the people responsible for the intranet, can we see what questions are being asked and who is asking them?
Currently, as administrators, you can view questions and conversations if a user submits feedback. This gives you access to the entire conversation and information about the sources used in the response.
Can the assistant be positioned as a floating box at the bottom, as chat assistants usually are? Or does this require further development?
Currently, the Assistant can be displayed as a search box or as a button, with a few different appearance settings. To make it float at the bottom, like a classic chat assistant, you need to add some CSS of your own. However, this is a very good suggestion for improvement and something we may add as a future setting.
My personal information, such as vacation days, requires integration with the HR system, correct? How does this function from an information security perspective?
Currently, Sitevision primarily handles more traditional information, such as pages, articles, and documents. However, our roadmap includes support for integrations with other systems, such as HR systems, in order to display real-time data such as the number of vacation days for each individual user.
How does this solution handle different languages?
The assistant and its AI are very good at handling different languages. This means that a user can, for example, ask a question in a language other than the one the page is written in and still get a correct and translated answer back.
If you want to connect external services, does the assistant cost extra?
No, there is no additional charge. The ability to connect external services is included in the Assistant add-on.
Can the assistant limit responses based on the user's target audience?
Yes, the Assistant takes permissions into account. This means that if, for example, you have a manager login that only managers have access to, a user without permissions will not be able to receive answers based on the content there.
In addition, the Assistant can also be used in conjunction with target groups and views in Sitevision, making it possible to further customize responses for different user groups.
Do you pay per token regardless of whether it is an external or internal solution?
No, you don't pay per token. Sitevisions Assistant is offered at a fixed monthly cost, making it easy to predict and plan your costs regardless of whether the solution uses internal or external sources.
Is it possible to have the assistant expand on the page surface instead of as an overlay?
Not at present, but the plan is to add more appearance settings in the future.
Is the assistant accessible?
Yes, the Assistant is accessible. The module has been tested for accessibility, just like all our other modules, to ensure that it meets current requirements and works for all users.
Do you have any restrictions preventing visitors from entering personal information in the fields?
Nej
In which SV version are the AI assistants available?
The assistant requires at least version 2025.07.1
Will it be possible in the future to have a conversation saved between uses/sessions, so that you can keep a history of what you have asked?
There is a plan to enable conversations between sessions to be saved in the future. This means that users will be able to keep their history and easily go back to previous questions and answers.
Can the assistant completely replace the search module?
The search module and the Assistant fulfill different needs and therefore complement each other. The search module is best when you want to quickly find a specific page or document, and it can also list content so that the user can get an overview and choose what is most relevant.
The Assistant can provide more explanatory and coherent answers by combining information from multiple pages, articles, and documents. Therefore, I would not say that the Assistant can completely replace the search module, but rather that it serves as a valuable complement.
Is it possible to automatically retrieve questions for the conversation starter from the chats as they happen? (I.e., real "frequently asked questions.")
Not at present. Currently, the questions in the conversation starter are editorially controlled, meaning that you choose which questions are displayed. However, this is a very good suggestion for improvement that we will take into consideration going forward.
How does it handle different versions of the same page where one may be more current than the other?
If the Assistant finds conflicting information, it may have difficulty providing a clear answer. Think of the Assistant as a regular user: if a user encounters two different versions of the same page with different content, it will also be confusing.
Therefore, it is important to keep the content up to date and avoid duplicate or conflicting versions of the same information. This ensures that the Assistant can always provide answers that are as accurate and clear as possible.
How can you protect yourself against security threats linked to AI, such as malicious input on external websites?
The assistant is protected in the same way as other functions in Sitevision, such as search. This means that it is covered by the same security layers and protection against malicious input as other parts of the platform. This minimizes the risk of attacks or improper use, even when the assistant is exposed on external websites.
Is it possible to exclude certain pages and control which ones the AI should search for information on?
Yes, you can control what information the Assistant has access to. This means that you choose which pages, files, and articles to include in the Assistant's knowledge base and, consequently, which ones to exclude. This gives you complete control over the content used to answer questions.
And it is not trained on our data, but only uses it as a "temporary source." None of our data leaves us?
No, your data is not used to train the models. All data is stored in Swedish data centers, and no general model training is performed on customer information.