The main chat interface is the primary way of interacting with Agent. From here you can start and manage conversations, use your own files or external sources as context, and switch between text, speech‑to‑text, and voice chat.

Each chat combines three elements:

• A model that determines how Agent generates answers.

• An optional instruction that defines how Agent should behave (tone, role, level of detail).

• An optional context – Sources and Resources (file groups or uploaded files) you select to inform Agent’s responses.

The sections below describe the main chat features, where they are found and how they can be adjusted. Advanced actions like data analysis, translation, and file processing are covered on dedicated pages.

• With default settings: Click Chat in the left-hand menu. This opens a new conversation with the default Model and no Instruction, Resources or Sources selected. Use this when you’re starting a completely different task.

• With current settings: Click the Reset Chat button (top right). This opens a new conversation that uses the same Model, Instruction, and selected Resources or Sources as the current chat. Any text in the Chat Input field will be retained. Use this when you want to start a fresh thread but with the same configuration.

Open

Reset Chat

• Conversation area: Shows your messages and Agent’s replies in order. New messages appear at the bottom. Click a message’s ⋮ menu button to copy it or exclude it from context (see Response Details ).

• Chat Input: Located at the bottom of the interface; type your prompts, questions, or commands here. You can also paste or drag supported file types directly here for use in the current conversation.

• Conversation History: Scroll up to review previous messages at any time.

At the top of each AI reply is a bar containing several items:

Open

• Answer: The main response to your prompt.

• References: If any Resources or Sources were used to produce this response, they are shown here. Clicking the References tab opens a detailed list of Resources. If the Web source was used, the tooltip and dialog also show the Web documents used to generate the answer.

• Steps:

  • If /reason was used (see Agent - ReasonPreview ), this section shows the reasoning steps used by the Agent.

  • If Sources were used, this section shows the information retrieved from them.

• Model name: The name of the model used to generate the response.

• Instruction name: If an Instruction was used, this will show its name.

• Menu button: Opens additional options:

  Open

  

  • Copy: Copies the response using Markdown to preserve formatting

  • Exclude from context/Include in context: Toggle whether this message is included in the ongoing chat context used by the model. Use this if a particular prompt or reply is confusing the model and you want the model to ignore it.

  • Convert to document: Generates a Word .docx document containing the response body

  • Convert to Presentation: Generates a PowerPoint .pptx document containing the response body

Open

• Model Selection: Select from the AI models available in your Agent instance. Different models may offer different capabilities or trade off speed, cost, and quality. Choose a model suitable for your task or follow your organization’s guidelines. In some types of Agent deployments, the set of available models can be modified by the application’s administrator.

• Instructions (formerly Personalities): Define how Agent should behave in this chat (for example, as a proofreader, analyst or IT professional) See Instructions

Open

• Suggestions: When enabled, Suggestions propose text completions based on your previous prompts as you type, which can speed up repetitive requests.

You can insert specialized actions via the Actions button or by typing /. Once selected, the action command is inserted at the beginning of the prompt. Use them when performing a specific task such as data analysis, translation, or file processing.

Open

Open

• /analyze: Perform data analysis and generate visualizations from uploaded files or direct data input. See Agent - Advanced Data AnalysisPreview for details.

• /generate-image: Create an image from a text description (for example, a simple diagram or illustration)

• /reason: Decompose complex questions into subproblems and return a structured, well-reasoned answer. See Agent - ReasonPreview for details.

• /translate: Translate files from one language to another. See Agent - TranslationPreview for details.

• /process-file: Process the selected file according to your instructions. See Agent - Process FilePreview for details.

Agent supports Speech-to-Text and Speech-to-Speech, enabling interaction via voice as well as keyboard.

• Speech-to-Text: Dictate requests using your microphone.

• Speech-to-Speech: Conduct a spoken conversation with Agent.

See Agent - Speech FeaturesPreview for details on their usage, requirements and browser support.

• Drag and Drop Files: You can upload files directly into the conversation. Images can be dragged into the Chat Input field. If the /analyze action is active, you can also drag spreadsheets there. Other file types must be dropped into the Resources panel, where they are processed. See Uploading Files for more ways to upload files to Agent.

• Selecting Files or File Groups: In the Resources panel, use the checkboxes next to files and file groups to include them as context in the conversation.

  • Checking a box includes that item; unchecking removes it from context.

The panel on the right side of the screen lets you choose which external Sources, uploaded files, and File Groups Agent can use as context when answering your questions.

This panel can be collapsed for a clearer layout and expanded when you need to change context. For a detailed overview of the panel’s functionalities, see Agent - Sources and Resources panelPreview .

Open Sources/Resources panel (collapsed)

Open Sources/Resources panel (expanded)

LLMs often have limitations on the maximum length of their responses, and will cut off the response when this limit is reached. In some of the LLMs, such as Claude Opus, telling the LLM to Continue will let it continue from where it left off.