Attachments
You can attach files to your messages so the agent can read, analyze, or reference them. Attachments are scoped to the conversation where they are uploaded.
Supported file types
CubePlex classifies attachments into three kinds:
| Kind | Examples |
|---|---|
| Image | PNG, JPEG, WebP, GIF |
| Document | PDF, plain text, Markdown, CSV, DOCX, XLSX, JSON, YAML |
| Other | Any other file type allowed by your organization's configuration |
The exact set of allowed MIME types is configured by your org admin. If you try to upload a file type that is not allowed, the upload is rejected with an error.
How to attach files
Click to upload
- Click the paperclip icon in the input bar.
- Select one or more files from the file picker.
- The files appear as chips above the input area while they upload.
- Type your message (optional) and press Enter to send.
You can attach multiple files to a single message, up to the per-message cap (see Size and quota limits).
Drag and drop
Drag files from your desktop onto the chat area. A drop zone overlay appears confirming the drop target. Release the files and they begin uploading immediately.
Remove before sending
While files are still uploading or staged (before you send the message), click the X on any file chip to remove it. Once a message is sent, its attachments become permanent and cannot be removed.
How attachments reach the agent
When you send a message with attachments, the files are:
- Stored in CubePlex's object store.
- Injected into the agent's sandbox at a known path so it can access them with code execution tools.
- Referenced in the message context so the agent knows what files are available and their types.
For images, the agent can also "see" the image directly through vision-capable models.
Viewing attachments in the chat
Sent attachments appear above the message bubble:
- Images display as thumbnails. Click a thumbnail to open a full-size lightbox.
- Documents and other files display as compact chips showing the filename and size. Click a chip to open a preview in the side panel:
- PDF, Markdown, plain text, code, JSON, and CSV files render directly.
- Word, Excel, and PowerPoint files (
.docx/.xlsx/.pptx) render through Microsoft's Office Online viewer. This requires the deployment to have a public URL configured (api.public_url); the viewer fetches the document through a short-lived tokenized link. - Audio and video files open with playback controls.
- File types that can't be previewed (archives, unknown formats) show a download button instead.
Size and quota limits
| Limit | Default |
|---|---|
| Max file size | 50 MB per file |
| Max files per message | 10 attachments referenced in a single message |
| Max per conversation | 500 MB total across all attachments |
These limits are configurable by the org admin. If you exceed the per-file limit, the upload is rejected. If you attach more than the per-message cap, sending the message is rejected — split the files across multiple messages. If you exceed the per-conversation quota, you need to start a new conversation or ask an admin to adjust the limit.
Tips
- Attach context documents early. Upload reference material (specs, data files, style guides) in your first message so the agent has full context from the start.
- Use images with vision-capable models. If you need the agent to interpret screenshots, diagrams, or photos, make sure the selected model supports vision. See Model Selection.
- Combine text and files. Always include a text message explaining what you want the agent to do with the attached files. A bare file upload with no instructions produces less useful results.
- Check file size before uploading. Large files (especially images) can slow down the upload. If you are sharing code or data, consider zipping multiple files into a single archive.