Skip to content

Administrator Features

This section describes the features available to administrators of the AI Support appjar. Administrators are responsible for configuring and maintaining all the components that power the AI assistant experience. These management views are accessible through the AI Support section in the application sidebar.

The administrator features are organized around the following areas:

  • Assistants: Create and manage AI assistants by combining a language model, a prompt, and document categories into a cohesive configuration. Assistants are what users interact with in the Chat view.
  • Documents: Upload and manage the knowledge base documents that assistants use for Retrieval-Augmented Generation (RAG). Documents are processed into embeddings that the AI uses to find relevant context when answering questions.
  • LLMs: Configure connections to large language model providers: OpenAI, Anthropic, Gemini, and Ollama. Each LLM connection defines the endpoint, credentials, model, and the kinds of attachments the model accepts.
  • Prompts: Write and manage the prompts that define the behavior and personality of AI assistants. Prompts are the instructions given to the language model, after the rules every assistant follows, that shape how it responds.
  • Categories: Organize documents and assistants into logical categories. Categories allow administrators to control which documents an assistant has access to, enabling fine-grained knowledge scoping.
  • LLM Inspector: Review the detailed exchanges between the system and the language model. The inspector provides visibility into the requests sent to the LLM and the responses received, which is valuable for debugging and quality assurance.
  • Channels: Configure the messaging integrations (WhatsApp, Facebook, Instagram, and web channels for the embedded chat) through which the AI assistant communicates with users outside the application. Each channel links a platform account or a website to an AI assistant and includes webhook setup for Meta platforms.
  • Setting up Meta Channels: A step-by-step guide to connecting WhatsApp, Facebook Messenger, and Instagram through a Meta app.
  • Setting up the Embedded Chat: A step-by-step guide to showing the chat bubble on another website, where anonymous visitors talk to an assistant.

Filtering the Lists

The Assistants, Documents, LLMs, Prompts, Categories, and Channels views show their filters in a bar above the grid, followed by the view's own actions, such as its create button. While the filters fit in two rows they stay inline, next to the Filter button and an icon-only Clear button. Both buttons become active as soon as any filter field has a value, and pressing Enter in any field applies the filters as well.

When the filters would need more rows, as on a narrow window, they collapse into a single Filters button. A badge on it counts the active filters, and clicking it opens the Configure filters dialog, which holds the same fields with its own Filter and Clear buttons.

Using the Views on a Phone

The management views adapt to small screens. On a phone, their grids hide some of their columns and keep the actions menu at the end of each row; tapping a row shows the values of its hidden columns beneath it. Dialogs open full screen.

The phone's Back closes an open AI Support dialog before it leaves the page. A dialog with unsaved changes still asks whether to discard them, and a dialog opened from another one returns to it, as its Go back button does. Browsers that do not support this, such as Safari, keep their usual Back behaviour.