Platform Overview

The IllumiDesk Platform consists of an Admin site, End-User interfaces, and the REST API.

1. The IllumiDesk Admin, located at app.illumidesk.ai, site is where users can create and manage data sources, assistants, publish assistants using integrations, and analytics.

2. The IllumiDesk End-User interfaces (IllumiDesk includes a default interface locate at app.illumidesk.ai/chat as well as embeds or integrations for third party applications such as Slack, is where end-users can interact with assistants configured by account managers with admin privileges.

3. The IllumiDesk REST API is where more technical users can integrate configured assistants into their existing stacks or create their own chat interfaces.

As you use IllumiDesk, you and your team will interact with all of these parts. Here is a diagram that helps illustrate the IllumiDesk parts mentioned above:

You may notice more details with this diagram, so let’s dive into some of the concepts:

• Data Sources: these are the locations and connection properties for how you, as an Admin user, can configure connections to a variety of data sources with a few simple clicks. These options include Files, Sharepoint, Confluence, Website URLs, Google Drive, among many others.

• Vector Store: the vector store is where the files are stored and indexed once they are synced with your data sources. Due to the nature of how Large Language Models (LLMs) work, data is first converted into chunks of text, scored with an “embed” model, and stored/indexed in a database. Vector stores are used to improve the quality of the AI’s responses without having to go through expensive and time consuming model training jobs that “bake in” the content into the model itself.

• Assistants: assistants are instances of an assistant that include instructions, knowledge bases associated to data sources, and where they are surfaced for end-users. One IllumiDesk account can have many assistants, one assistant can have many chat threads. and one thread can have many messages.

• Assistants have memory, agents, and logs. Memory is used to “remember” end-user preferences within a thread with one or more messages. Memory is also useful to help improve the assistant’s response quality over time by continuously feeding back the assistant whether or not the response was accurate, whether or not the response fit the brand’s tone, etc. Finally, logs are used to trace how the assistant behaves over time. Logs are useful for reliability, understand where there may be opportunities for improvement, and analytics.

• Agents are used to retrieve context either from the Knowledge Base or from third party systems in real time, and, potentially, to take action on if required.

The next sections will put these concepts into practice to help you build and assistant using custom data sources, share your assistant with end-users, and track it’s use over time.