@attribute [Route(Routes.ASSISTANT_ERI)] @using MudExtensions @inherits AssistantBaseCore You can imagine it like this: Hypothetically, when Wikipedia implemented the ERI, it would vectorize all pages using an embedding method. All of Wikipedia’s data would remain with Wikipedia, including the vector database (decentralized approach). Then, any AI Studio user could add Wikipedia as a data source to significantly reduce the hallucination of the LLM in knowledge questions. Related links: ERI repository with example implementation in .NET and C# Interactive documentation aka Swagger UI
Auto Save The ERI specification will change over time. You probably want to keep your ERI server up to date. This means you might want to regenerate the code for your ERI server. To avoid having to make all inputs each time, all your inputs and decisions can be automatically saved. Would you like this? Common ERI server settings @foreach (var language in Enum.GetValues()) { @language.Name() } @if (this.selectedProgrammingLanguage is ProgrammingLanguages.OTHER) { } @foreach (var version in Enum.GetValues()) { @version } Download specification Data source settings @foreach (var dataSource in Enum.GetValues()) { @dataSource.Name() } @if (this.selectedDataSource is DataSources.CUSTOM) { } @if(this.selectedDataSource > DataSources.FILE_SYSTEM) { } @if (this.NeedHostnamePort()) {
@if (this.dataSourcePort < 1024) { Warning: Ports below 1024 are reserved for system services. Your ERI server need to run with elevated permissions (root user). }
} Authentication settings @foreach (var authMethod in Enum.GetValues()) { @authMethod.Name() } @if (this.selectedAuthenticationMethods.Contains(Auth.KERBEROS)) { @foreach (var os in Enum.GetValues()) { @os.Name() } } Data protection settings @foreach (var option in Enum.GetValues()) { @option.Name() } Embedding settings You will likely use one or more embedding methods to encode the meaning of your data into a typically high-dimensional vector space. In this case, you will use a vector database to store and search these vectors (called embeddings). However, you don't have to use embedding methods. When your retrieval method works without any embedding, you can ignore this section. An example: You store files on a file server, and your retrieval method works exclusively with file names in the file system, so you don't need embeddings. You can specify more than one embedding method. This can be useful when you want to use different embeddings for different queries or data types. For example, one embedding for texts, another for images, and a third for videos, etc. Name Type Actions @context.EmbeddingName @context.EmbeddingType Edit Delete @if(this.embeddings.Count == 0) { No embeddings configured yet. } Add Embedding Data retrieval settings Provider selection for generation