Skip to main content

AI Chat

The AI Chat is your primary interface for conversational AI on the platform. Use it to ask questions, search your organization's knowledge base, pull in live web results, analyze uploaded files, and have multi-turn conversations powered by large language models.

Getting Started​

Open Chat from the sidebar. The first time you visit, the page shows an empty state with a prompt to create your first session. Click New Session in the sidebar or the main content area to begin.

Each session is an independent conversation thread with its own history, model selection, and feature toggles. You can have multiple active sessions and switch between them at any time.

Interface Overview​

The chat page is divided into two areas:

AreaDescription
Session sidebarLists all your sessions with search, status filters, and quick actions
Conversation panelShows the message thread, input controls, and streaming responses

On mobile, the session sidebar opens as a slide-out drawer from the menu icon.

Header Controls​

Above the conversation you find:

  • Session title — the current session name (click to expand details)
  • AI model selector — switch the language model for this session
  • Knowledge Search toggle — enable or disable knowledge base retrieval
  • Web Search toggle — enable or disable live web results
  • Persona badge — shows which persona is active, if any
  • File uploads indicator — appears when file attachments are enabled
  • Pin header — keep the header visible while scrolling through long conversations

Sessions​

Creating a Session​

Click New in the session sidebar to create a session instantly. The platform assigns default settings from your tenant configuration and any persona assigned to you.

Session List​

The sidebar shows your sessions sorted by most recent activity. You can:

  • Search — type in the search box to filter by title
  • Filter by status — switch between All, Active, Completed, and Archived
  • Select — click a session to load its conversation
  • Edit — open the overflow menu (three dots) to rename, change settings, or update the description
  • Archive — move finished sessions to the archive for a cleaner list
  • Archive all completed — bulk-archive every session in the Completed status

Session Settings​

Open the Edit Session dialog from the overflow menu to adjust:

  • Title and description
  • Status — Active, Completed, or Archived
  • Knowledge Search — enable or disable knowledge base retrieval
  • Web Search — enable or disable live web results
  • File Uploads — allow or disallow file attachments
  • AI Model — choose from the available models when your tenant has more than one

Session Lifecycle​

Sessions move through three statuses:

  1. Active — the session is open for messages
  2. Completed — the session has been closed (manually or automatically after a period of inactivity)
  3. Archived — the session is stored for reference but no longer active

Sending a message to a completed session automatically reactivates it.

Tip: Your tenant may limit the number of concurrent active sessions. When you reach the limit, the oldest active session completes automatically to make room.

Sending Messages​

Type your message in the text field at the bottom of the conversation panel and press Enter or click the send button. While the AI processes your request you see a streaming response card that renders tokens as they arrive.

Streaming Responses​

Responses stream in real time. A green dot in the header indicates a live connection. You can:

  • Stop generation — click the stop button on the streaming card to halt the response early
  • View sources — click the knowledge or web source chips on the response to open the Sources Drawer

Markdown and Citations​

AI responses render full Markdown including headings, lists, code blocks, and tables. When knowledge or web sources inform a response, numbered citation chips appear inline. Click a citation to scroll to the corresponding entry in the Sources Drawer.

When Knowledge Search is enabled for a session, every message you send also queries your organization's knowledge base. The AI augments its answer with relevant documents and surfaces them as citations.

How It Works​

  1. Toggle Knowledge Search on in the header controls or session settings
  2. Send a message
  3. The platform searches indexed documents, respecting the categories and indices accessible through your persona
  4. Relevant excerpts appear as knowledge sources in the response
  5. Click a citation chip or the Sources button to browse full results in the Sources Drawer

Sources Drawer​

The Sources Drawer opens as a side panel and lists every knowledge source and web result for the current response. Each entry shows the document title, a relevance snippet, and a link to preview the original file.

Tip: Knowledge search respects your persona's category and index access. If you cannot find a document you expect, ask your administrator about your persona's scope.

When Web Search is enabled, the AI supplements its answer with live results from the web. This is useful for questions about current events, recent data, or topics outside your knowledge base.

  1. Toggle Web Search on in the header controls or session settings
  2. Send your message
  3. Web results appear alongside knowledge sources in the Sources Drawer

Web search usage may be subject to quota limits configured by your administrator.

File Attachments​

If file uploads are enabled for your session, you can attach documents for the AI to analyze within the conversation.

Uploading Files​

  1. Click the attachment icon next to the message input
  2. Select a file from your device (allowed types depend on your tenant configuration)
  3. A chip appears above the input showing the file name and processing status
  4. Once processing completes, send your message — the AI includes the file content in its analysis

File Status​

Uploaded files pass through several stages:

StatusMeaning
UploadingThe file is being transferred
ProcessingThe platform is extracting and indexing content
ReadyThe file is available for the AI to use
ErrorSomething went wrong — check the error message

Attachments are temporary and expire after a configured period. They are not stored permanently in your Files library.

Important: There is a per-session quota for file attachments. A warning appears near the input field when you approach the limit.

Personas​

Personas define the AI's behavior, knowledge scope, and available features for a session. Your administrator assigns one or more personas to you, and each persona controls:

  • Which AI model to use by default
  • Which knowledge categories and indices the AI can access
  • Whether knowledge search, web search, and file uploads are available
  • Default retrieval settings such as the number of documents and relevance threshold

When you create a session, the platform applies the settings from your assigned persona. If you have multiple personas, contact your administrator about which one is active for new sessions.

Note: Session-level toggles (Knowledge Search, Web Search, File Uploads) can only be turned on if the persona and tenant configuration allow it. You cannot enable a feature that your persona restricts.

Customer and folder scope​

When Customers is enabled, you can narrow knowledge search:

  • A persona may allow all customers (including files not linked to any customer) or a fixed allowlist
  • The session customer picker is a further subset. If nothing remains after intersecting with the persona, search returns nothing rather than leaking
  • The picker is hidden when the session is already locked to one customer (the customer Ask tab)
  • Optional folder selection further restricts by physical folder membership — that is location, not customer folder designation

Link files to customers in Files or on the customer Files tab so scoped chat has something to retrieve.

Chat Analytics (Administrators)​

Administrators see an Analytics button in the session sidebar. The analytics view shows usage metrics for the current stack:

  • Overview — total sessions, messages, token consumption, and average session duration
  • LLM Model Usage — breakdown by model
  • Web Search Usage — query counts and provider distribution
  • Activity — trend charts over the selected time range (7, 30, or 90 days)

Mobile Experience​

On phones and small tablets:

  • The session sidebar opens as a drawer from the menu icon in the top bar
  • A compact session selector lets you switch sessions quickly
  • The header controls stack vertically to fit narrower screens
  • All core features (messaging, knowledge, web search, files) work the same way

Names and identifiers sent to language models​

A tenant administrator can turn on Pseudonymise names and identifiers sent to language models under Settings → AI Model Configuration. The switch is off until someone turns it on. It applies to chat and to other language-model features, including agents.

While it is on, direct identifiers are replaced before any text is sent to an external language model. The answer you read has the real values restored.

What is replaced:

  • Client and contact names, email addresses, phone numbers, and tax or registration numbers stored on Customers
  • Email addresses, phone numbers, IBANs, and Croatian personal identification numbers (OIBs) found in the text

The lookup that maps tokens back to those values stays in your environment. It is used only to restore the answer. It is not sent with the prompt.

Also while the switch is on:

  • Image attachments on a message are not sent to the model
  • Documents already stored in Files, citations, and knowledge search stay as they are
  • Scanned-document OCR and the search index are unchanged

Important: This reduces direct identifiers. It does not make a prompt anonymous. A role and a place with no name attached can still be in the text. Some workflows also withhold a whole section, such as a customer or supplier list or an internal grade, on the steps that do not need it.

Best Practices​

  • Be specific — detailed prompts produce better answers, especially when knowledge search is on
  • Use knowledge search for internal questions and web search for external or time-sensitive topics
  • Organize sessions by topic — give each session a descriptive title so you can find it later
  • Archive regularly — move completed sessions to the archive to keep the sidebar focused
  • Review sources — always check the Sources Drawer to verify the information the AI cites

Troubleshooting​

The AI response seems incomplete​

Problem: The response cuts off mid-sentence. Solution: This usually means the model's token limit was reached. Try rephrasing with a shorter prompt or switch to a model that supports longer outputs.

Knowledge search returns no results​

Problem: You enabled knowledge search but the response has no citations. Solution: Verify that your persona has access to the relevant categories and indices. Also confirm that the documents you expect are indexed in the knowledge base.

File upload fails​

Problem: The attachment shows an error status. Solution: Check that the file type is in the allowed list and the file size is within the configured limit. Try a different format if the issue persists.

An attached image was not described​

Problem: You attached an image and the reply does not discuss it. Solution: If Pseudonymise names and identifiers sent to language models is on, image attachments are withheld from the model. Turn the switch off for that conversation, or describe the image in text.

Cannot create a new session​

Problem: The platform shows a quota or limit message. Solution: You have reached the maximum number of concurrent active sessions. Archive or complete an existing session to free a slot.

⏱️ Read time: 14 minutes | 📊 Difficulty: beginner | 🔄 Last updated: 2026-09-26