Skip to main content
In Cloosphere Chat you can attach files for the AI to analyze, or connect a Knowledge Base for document-grounded answers.

​
Attaching Files

You can attach files in several ways.
Input area with attached files

Input area with attached files

Drag files into the chat area to upload them automatically. You can drag multiple files at once.

​
Supported File Formats

When you upload audio (WAV, MP3, etc.), STT (Speech-to-Text) is run automatically to produce a transcription.

​
File Size Limits

The maximum file size depends on the value the admin sets in Settings > Documents. Files exceeding the limit are rejected with a “File size should not exceed X MB” error.

​
Image Analysis

When you attach an image (PNG, JPG, GIF, WebP, AVIF), a Vision model analyzes its content.
Image analysis response

Image analysis response

  • Charts/graphs: Extract and analyze data points
  • Document images: OCR-based text extraction
  • UI design: Design feedback and code conversion
  • Error screens: Analyze error messages and suggest fixes
  • Photos: Object recognition, scene description
The product has no per-model image-support indicator. Picking a model without Vision can cause the image to be silently ignored, so switch models and retry if it isn’t recognized. The “Selected model(s) do not support image inputs” error appears only when no model is selected at all.
Enable Settings > Interface > Files > Image Compression to auto-compress high-resolution images to a target size, improving upload speed and token efficiency.

​
Document File Processing

For non-image files (PDF, DOCX, code files, etc.), text is extracted automatically at upload time.
1

Upload the file

Attach a file using one of the methods above. During upload it shows an “uploading” state.
2

Text extraction

The server extracts the file contents automatically. Once complete, the state changes to “uploaded”.
3

Use the context

The extracted text is included as context in the AI conversation, enabling questions about the file content.

​
Knowledge Base Integration (RAG)

​
Reference a Knowledge Base with #

Type # in the input box to see available Knowledge Base collections and individual files. The AI searches documents from the selected Knowledge Base or file and uses them in its answer.
Knowledge Base autocomplete dropdown

Knowledge Base autocomplete dropdown

​
How RAG Works

1

Query analysis

A search query is generated from the user’s question.
2

Vector search

Vector similarity search runs against document chunks stored in the Knowledge Base.
3

Build the context

The retrieved relevant chunks are inserted into the AI prompt as context.
4

Generate the response

The AI generates an answer that references the retrieved documents and provides citations.

​
Agent-based RAG

When an agent has Knowledge Bases attached, it searches them automatically — no # command required. Attach Knowledge Bases in Workspace > Agent settings.

​
Citation Display

Responses that referenced Knowledge Base or web search results show citations.
Citation display

Citation display

​
Citation List

An N Sources toggle appears below the response. Expand it to list the referenced sources as citation chips.
Sources that have a URL (web search results, for instance) render as an external link chip with a favicon and domain. Clicking one opens the original page in a new tab.

​
Reviewing the Excerpt (citation chip popover)

Hover over or click a citation chip to open a popover showing an excerpt of the retrieved chunk.
The popover shows an excerpt of the chunk, not the full source. Chunks longer than 1200 characters are truncated, and there is no screen that opens the whole original document. To read the full content, open the original file from Knowledge Base.
Only one popover is open at a time. Pointing at another chip closes the previous one automatically.

​
Files in Chat Controls

The Files section of the Chat Controls panel lists files attached to the current conversation.
  • Each item shows filename, type, and size
  • Remove individual files with the X button on the right of each item
  • Clicking a file shows its details