Do you find the diagram below a bit too technical? No problem! As an IT architect, it is my duty to provide a clear overview of how things work. Some of you will definitely appreciate this. Not one of them? Then take a look at the next section, where I explain how the multi-model concept works and how you can assemble your own team of AI models.

Authentication for BizEye is handled via Azure Entra ID. This means that only users known within your organization can gain access to BizEye. Further configuration within Azure can regulate this in an even more granular way, if desired. The Administrator determines which roles are set up in BizEye and which user is assigned to which role. A role has a set of assigned rights, or privileges. These rights determine access to the various functionalities of the application. The Administrator is the person responsible for configuring memory storage. Memory storage refers to an external source where data is stored and accessed through BizEye. This can involve structured data in a database or unstructured data from a set of documents. Through the configured roles, it can be determined who has access to which memory storage. If structured data is involved, access can be configured down to the entity level and even to the field level. This results in a highly flexible way to make your data generally available in BizEye while ensuring it is only accessible to users with the appropriate role.

The Administrator has the ability to configure multiple connections in BizEye. This enables BizEye to communicate with an external service. These connections are then used when setting up memory stores. By separating these two concepts, one connection can be used by multiple memory stores, with role assignments ensuring that some roles can see certain things while others cannot. This creates tremendous flexibility with minimal effort.

The list of connections continues to grow. Below is an overview of the connections currently available:

AI connections This type of connection forms the foundation for the proper functioning of BizEye. An OpenAI connection requires a subscription with OpenAI, whereas Azure OpenAI requires a subscription with Microsoft. OpenAI limits you to models from the well-known ChatGPT family, while Azure OpenAI provides the advantage of access to a long list of available AI models, including OpenAI’s models. If you know that a model is available via open source, there is a high probability that it is offered via Azure OpenAI in combination with Azure AI Foundry. Feel free to check the model catalog on Microsoft’s website. Another major difference between OpenAI and Azure OpenAI is the pricing model. The former charges a fixed monthly fee, whereas Microsoft charges based on usage. A hybrid architecture, where you can plug in your own AI models, is also possible. BizEye requires the following types of AI connections to function properly:
Semantic memory Semantic memory refers to the ability to store and retrieve knowledge in a way that understands the meaning of words, sentences, or documents. Instead of remembering exact words, AI understands the relationships between concepts, allowing it to find relevant information even when different words are used. This is done using embeddings, which convert text into numerical representations that capture meaning. BizEye requires the following types of semantic memory connections, at least one of which is necessary for proper functioning:
Entity memory Entity memory refers to the ability to store and retrieve structured data about specific entities, such as customers, products, or employees. Each entity has defined fields (e.g., name, ID, date of creation), and relations to other entities (orders are made by a customer, songs belong to an album) that allow for organized and efficient data management. Currently, entity memory access supports various data storage solutions: Security considerations are paramount when accessing entity memory. Each memory store's schema is stored separately within BizEye, granting administrators the ability to finely control access to specific entities, fields, and even rows of data. This ensures data security and confidentiality on the most granular level. Combined with the functionalities of the Excel plugin the options become unlimited.

Other In addition, there are several other types of connections. Just like for entity memory, they determine whether certain plugins can be used or not. Only Document Storage is required for BizEye to function properly:

The Administrator has the ability to configure multiple AI models in BizEye. As soon as multiple models are present in BizEye, they start forming a team together. Your chat session begins with an AI model set by the Administrator. If this AI model determines that certain tasks can be handled better by its teammates, it will autonomously seek advice from the rest of its team and coordinate everything. This all happens behind the scenes, without you having to do anything special! Some models excel at reasoning but, for example, cannot search the internet. Another model, however, can do that but cannot generate images. But don’t worry, there’s yet another AI model for that. All these models work together to ensure you get the answer to your question in the way you want. Here lies the true power of BizEye. As the AI models become smarter, BizEye will be able to take advantage of this!

Just a repetition of what is stated above. BizEye comes with a whole arsenal of plugins. Together with the team of AI models, you can achieve agentic AI—but in a way that you fully control. There are plugins to quickly make data available in Excel (without ever leaving your environment), plugins to automate repetitive tasks in Excel, plugins that make your solution smarter the more you use it, plugins that convert audio into text (think dictation), plugins that search the web, and even plugins that allow you to interact with your own file system on your PC or laptop.

The list of plugins keeps growing. Below is an overview of the plugins that are currently available:

Assistant knowledge This plugin allows the assistant to store general knowledge that may be useful for future interactions. It is designed for broad, reusable information rather than user-specific details. Administrators can manage memory stores, but users and stored data can be removed, so structuring stored knowledge effectively is important.

Memory store knowledge A memory store is a specialized feature within the application that extends the assistant's knowledge. Administrators can configure memory stores to provide domain-specific, company-specific, or topic-related information. This feature helps the assistant retain relevant knowledge about different memory stores and distinguish between general, user-related, or memory store-specific information. It can for instance be used to store certain concepts or terms for reusability.

User knowledge This plugin is used to store details about individual users, such as their preferences, background, or other relevant information. If a user requests that specific information be remembered for future interactions, this functionality ensures that their request is honored. This will give BizEye a personal touch for users who are interacting with it.

Embedding memory This plugin provides access to a semantic memory store, which holds unstructured data. It enables retrieval-augmented scenarios where user queries trigger searches within stored information to enhance responses with relevant details. Also all chat sessions of all users will be embedded and will only be accessible for those users. This enhances the personal touch for users, since the assistant is capable of remembering previous chat sessions.

Entity memory This plugin grants access to a structured memory store containing organized data such as tables, fields, and relationships. Unlike unstructured memory retrieval, this plugin enables structured queries similar to SQL, allowing for precise data retrieval. An extension to this is available to have BizEye create semantic relations between entities for you. Even when your structured data is not relational on a schema-level, this can be enforced simply be providing the right meta data in the structured memory store configuration.

DevOps This plugin integrates with an Azure DevOps instance, providing access to work items, sprints, and other project management data. It enables automated retrieval of relevant DevOps information through API calls.

Buddy service This plugin facilitates collaboration by connecting to a buddy service that assists with specific tasks or processes. This is what was discussed in the multi-model concept above and enables you to construct a whole "team of buddies" that can work together to answer a question."

Excel This plugin provides functions to interact with Excel files, leveraging the ClosedXML library, to perform various operations. Together with the entity memory plugin it will enable you to perform self-service BI on your structured datasets without the need to build Power BI reports on top of them. And the best thing of all is that no data will be passed to the AI model in any way, so there is no data in transit, meaning no risks for any data security breach at all.

Google search This plugin enables internet searches using Google Custom Search to retrieve information from online sources. This allows for accessing real time data from any source that is available on the internet.

Chat automation This plugin is designed to automate prompt handling in chat automation scenarios. It is typically used with a set of instructions provided to the assistant. The goal is to facilitate a structured question-and-answer process, ensuring that each instruction is properly addressed before proceeding to the next. This overcomes scenarios where a lot of data needs to be processed by the assistant, that might otherwise lead to the so-called context-memory-issue where there is a maximum amount of information that an AI model can process at a time. When a user requests automation for a task, this is the plugin that will be used.

Audio transcription This plugin provides the ability to transcribe audio files into text. Supported formats include WAV, MP3, OPUS/OGG, FLAC, WMA, AAC, ALAW, MULAW, AMR, WebM, M4A, and SPEEX.

Local file system This plugin enables interaction with the user's local file system. It operates through an asynchronous messaging system between the application and a locally running agent. It provides capabilities to read and create directories and files sitting on a user's PC or laptop. When combined with any other standard capability of an AI model, or any provided plugin, the sky becomes the limit.

Need a plugin that’s not on the list? Let us know!

Below is an overview of BizEye, purely from a technical perspective, focusing on the minimum required components. These components have been carefully selected to make BizEye as cost-efficient as possible—without sacrificing functionality or performance. Everything runs within your own Azure tenant, and except for the BizEye web app itself, nothing is exposed to the outside world. This ensures maximum functionality with minimal effort, while keeping security airtight.

Core components: