Click the New Chat button at the top of the sidebar, or use the keyboard shortcut Ctrl + Shift + O.The model selector button appears at the top left (header) and the input box in the center of the screen. Below the input box, a Suggestions area shows example questions registered on the selected model or agent (or the admin’s default examples if none are registered). Clicking a card fills the input box with that question — send it as is or edit it first.
2
Select an AI model
Click the model selector button at the top left to choose the model, agent, or flow to use. The button shows the name of the current selection (e.g. Cloosphere); it reads Select a model only when nothing has been selected yet.When the dropdown opens, there is a Search a model box and filter tabs at the top, and each list entry carries an Agent · Flow · Model badge and is grouped in that order. In other words, agents and flows registered in the workspace are chosen from the same dropdown as plain models. At the very bottom is a Temporary Chat toggle (it may be hidden depending on permissions) — turn it on and the conversation is not saved.
The model selector dropdown at the top
You can set a frequently used model as your default. On the new-chat screen, click the small “Set as default” button just below the model selector button and that model is selected automatically in future chats. (It sits in the header, not inside the dropdown, and does not appear once a conversation already has messages.)
3
Type your message
Type your question or request in the input box at the bottom of the screen.
4
Send the message
Press Enter or click the send button (up-arrow icon) at the right of the input box.To the left of the send button is a voice input (microphone) button that transcribes what you say, and while the input box is empty the send button’s position shows a voice button that opens a live voice conversation (it may be hidden depending on permissions).
Use Shift + Enter to insert a line break. Pressing Enter alone sends the message.
5
Read the AI response
The AI streams its response in real time, token by token, until generation completes.Click the stop button (square icon) during generation to cancel the response.
A response being generated. The far right of the input box becomes the stop button
Pick the model that best fits your task from the list registered by the admin. Below are typical model characteristics.
Model Type
Strengths
Recommended For
Large general-purpose model
High-performance multimodal model
Complex analysis, image understanding, coding
Lightweight / fast model
Fast and economical
General chat, simple Q&A, translation
Long-context model
Excellent long-context handling
Document summarization, report writing, analysis
Ollama models
On-premises execution
Sensitive internal data where security matters
The list of available models depends on admin settings, and the actual model names shown differ per instance. If a model you need isn’t visible, ask your admin.
Summarizing message bodies (attachments are not parsed)
Spreadsheets
XLSX, CSV
Data analysis, deriving insights
Images
PNG, JPG, GIF, WebP, AVIF
Image analysis, OCR (a vision-capable model is required — support isn’t shown in the UI, so switch models if the image isn’t recognized)
Code
PY, JS, TS, etc.
Code review, bug analysis, refactoring
Audio
WAV, MP3, etc.
Auto-transcribed to text on upload (STT)
Data
JSON, XML, YAML
Structured-data analysis
The actual allowed extensions can differ depending on admin settings.
Pair file attachments with a specific question to get more accurate answers.Example: “Summarize chapter 3 of the attached PDF and put the key conclusions in a table.”
When configured by the admin, you can pull files directly from Google Drive, OneDrive, or SharePoint. Choose the cloud storage option from the + button menu.
Generates a new response for the same prompt. After multiple regenerations you can switch between responses
View Trace
Opens the admin Tracing screen in a new tab to inspect this conversation’s execution path. Users without tracing permission see a Copy Message ID button in this position instead, which copies the Chat ID and Message ID for support requests
For responses containing code blocks, click the Copy button at the top-right of each block to copy just the code. Collapse/Expand and Save sit in the same spot; Python code additionally gets Run (when the admin has enabled code execution), and HTML/SVG code gets Preview.Select text inside a response and floating Ask and Explain buttons appear above the selection. Ask lets you write a follow-up question quoting the selected sentence, while Explain immediately asks for that part to be spelled out.
Cloosphere maintains conversation context. It remembers previous turns, so follow-up questions flow naturally.
[First question] Write Python code that calls a REST API[Follow-up] Add error handling[Follow-up] Convert it to an async version
As conversations grow long, the influence of earlier messages can fade. When the topic shifts, starting a fresh New Chat often produces more accurate answers.
Specific conditions yield more accurate results than vague requests.
Less effective
More effective
”Write a report"
"Write a Q1 2024 marketing performance report. About 2 A4 pages, including key KPIs and improvement areas."
"Fix this code"
"This Python code throws a TypeError. Find the cause and provide the fixed code.”
Specify the output format
Specifying the desired output format reduces post-processing time.
“Organize as a table”
“Write in markdown”
“Output as JSON”
“Explain step by step with numbered items”
Assign a role
Giving the AI a specific role yields more expert-level answers.
“You’re a senior Python developer. Review this code.”
“Analyze this campaign strategy from a marketing expert’s perspective.”
If you reuse the same role often, register it as an Agent so you don’t have to specify the role each time.
Break it into steps
For complex work, splitting into steps tends to produce better results than asking for everything at once.
Step 1: Find the top 3 trends in the attached dataStep 2: Analyze the cause of each trendStep 3: Organize the analysis into an executive briefing slide structure
