# How does askEdgi works? ## askEdgi Architecture

1. **User Query:** Everything starts with a simple question or query from the user. This could be about finding specific information, understanding a data object, or navigating the application’s platform on any term/object summary page. 2. **Similarity Search with VectorDB:** When the user clicks 'Enter,' AskEdgi gets to work. It first sends the query to a specialized database called VectorDB. Here, it performs a "similarity search." This search looks for templates predefined in each intent and information similar to the entered question. 3. **Matched Documents:** After the similarity search, AskEdgi now has a collection of matched documents. It thinks these pieces of information are most relevant to the query. 4. **Sending to OpenAI:** AskEdgi then sends the query and the matched content to OpenAI. OpenAI reads through this content to understand the context and find the most accurate and helpful response to the question. 5. **Accessing Proprietary Information:** If the query involves proprietary or specific information from the OvalEdge platform, OpenAI communicates with AskEdgi to fetch this data. This ensures that the user gets the most personalized and precise response possible. 6. **Elastic Search for External Data:** AskEdgi does not stop there. It also checks Elastic Search for any additional external documents or resources that could help answer the question better. 7. **Combining All Resources:** Once all the information is gathered—whether from VectorDB, OpenAI, or Elastic Search—AskEdgi combines these insights to craft a final response that is comprehensive and easy to understand. 8. **Delivering the Final Response:** Finally, AskEdgi presents the answer. It is like having a knowledgeable assistant who has reviewed all the relevant resources to provide the user with the best possible answer. ### Chat Interface The chat interface in askEdgi is designed to provide users with a smooth and user-friendly way to interact with the AI, search for metadata, and retrieve information. Below are the key components of the chat interface. #### System-Defined Intents System-defined intents are pre-configured intentions of actions that have a set of templates that guide users in their interactions with AskEdgi. These intents help the AI understand what the user wants to achieve and provide relevant templates and responses to the user.

#### Chatbox In the chatbox, users type their prompts or queries to interact with askEdgi. The chatbox will also suggest users' templates based on the words they type so that users can use the existing templates directly. The templates used in the chatbox can be modified and sent as a prompt to the AI to get accurate responses.

#### Response Section In the response section, askEdgi displays the results or answers generated in response to the user's prompt.

* **Output Formats:** Depending on the query, results can be displayed as text, tables, or hyperlinks. * **Interactivity:** Users can copy text, follow hyperlinks to detailed pages, or view tables within this section. ### Thread Management Thread management helps users organize and track their interactions with askEdgi over time.

* **Thread List:** Displays previous chat sessions (threads) on the left side of the interface. * **Functionality:** Users can create new threads, search for existing ones, rename or delete threads, and continue conversations. * **Highlighting:** The current active thread is highlighted to help users keep track of their ongoing sessions. ## Security askEdgi is available for all users (Viewers and Authors), but visibility and accessibility can be controlled through Application Security.

askEdgi's functionality can be restricted based on user roles or specific user settings. Administrators can configure access to askEdgi through application security settings, ensuring that only authorized users have access to this feature. * **Role-Based Access:** Only users with the appropriate roles can access askEdgi. This ensures that sensitive information is protected and only available to those with the necessary permissions. * **User Restrictions:** Specific users can be granted or denied access to askEdgi, allowing for specific control over who can use this feature. ## askEdgi Audits This feature is available for viewer users and authors, and end-users are concerned about the data or metadata being exposed to AI. Therefore, in the Audit Trials, we have divided the AI audit into two sections. 1. **AI Prompt -** Tracks the audit of descriptions generated by AI. 2. **askEdgi -** Tracks the audit of each user's interactions with askEdgi. Both audits show the metadata sent to OpenAI and the results received. The audit trails also show the number of tokens consumed during each execution.