Attaching Documents to Chat
You can attach a file directly to a chat message and ask the AI questions about its contents. Hubrix uses retrieval-augmented generation (RAG) to pull the most relevant parts of the document into the AI's context — so even a large file works well.
How to attach a file
There are two ways:
Option A — Drag and drop: Drag a file from your desktop directly onto the chat input area. A drop zone appears to confirm the file is ready.
Option B — Paperclip button: Click the paperclip icon in the chat toolbar. A file picker opens. Select your file and click Open.
The file name appears as an attachment chip above the message input. You can remove it by clicking the × on the chip before sending.
Supported formats
| Format | Extension |
|---|---|
.pdf | |
| Word document | .docx |
| Excel spreadsheet | .xlsx |
| PowerPoint | .pptx |
| Plain text | .txt |
| Markdown | .md |
| CSV | .csv |
| JSON | .json |
| HTML | .html |
Maximum file size for chat attachment is 50 MB. For larger files, upload them to Documents first, then use an agent or workflow with a RAG Query step.
How the AI cites document chunks
After you send the message, Hubrix:
- Splits the document into chunks of roughly 500 tokens each.
- Embeds each chunk as a vector.
- Finds the chunks most relevant to your question.
- Passes those chunks to the AI as context alongside your message.
The AI's response references the document text in its answer. Look for quoted text or phrases like "according to the document" — these trace back to specific sections.
The AI does not read the entire document at once. It reads only the most relevant chunks. For comprehensive analysis of every sentence in a long document, consider using Deep Research with the document uploaded to your workspace instead.
Tips for better results
- Ask specific questions — "What does section 3.2 say about payment terms?" works much better than "What does this say?".
- Text-based PDFs — scanned documents without OCR extract poorly. If your PDF looks like a photo of a page, the AI will have little usable text.
- One document per message — attaching multiple files in a single message is supported, but responses tend to be more accurate when focused on one document at a time.
Was this helpful?