RebelVault
RebelVault lets you upload personal working documents to RebelCore Agent and ask questions that can use those files as context.
RebelVault is separate from imported datasets. Dataset modules are governed through projects, semantic training, and exposure levels. RebelVault files are user-owned documents stored in your organisation’s tenant storage and searched only when your Agent setting allows file content.
Customer Super Admins can also publish organisation documents from Admin RebelVault. Those files are not personal uploads, but they can appear in Agent answers when they have been shared with you or with the whole organisation.
Upload A File
You can upload from the paperclip beside the chat box or from the RebelVault panel.
Supported formats:
| Format | Notes |
|---|---|
.txt / .text | Plain text files. |
.pdf | Text-based PDFs. Scanned image-only PDFs may not produce usable text. |
.docx | Microsoft Word documents. |
Maximum file size: 20 MB.
After upload, RebelCore extracts text, splits it into chunks, stores the original file in tenant blob storage, and queues an embedding job. The file may show as Embedding for a short period. Agent can search it after the status changes to Embedded.
Ask Questions
Open the settings cog and check Allow File Content when you want Agent to search your RebelVault files during a prompt.
When disabled, the vault_search tool is blocked for that request even if files exist. This is useful when you want a conversation to use only selected semantic datasets or general reasoning.
RebelVault search is scoped to:
- Files you uploaded.
- Files another user explicitly shared with you.
- Files shared with everyone in your organisation, including admin-managed RebelVault files.
Agent does not choose the tenant or user scope. RebelCore injects that scope server-side before the search tool runs.
Sharing
File owners can share a vault file with:
- A selected active user in the same organisation.
- Everyone in the organisation.
Shared users can search the file through Agent, but only the owner can delete the file or revoke shares.
Security Model
RebelVault uses the same customer boundary as the rest of RebelCore:
- Original files are stored in the customer’s own storage account.
- File metadata and chunks live in the customer’s tenant database.
- Search results are filtered by file ownership and active share grants.
- The language model cannot provide
customer_idoruser_id; those values are injected by RebelCore. - The Allow File Content preference can disable file search per user and per request.
Personal-information masking still applies according to your Agent privacy settings. RebelVault controls whether uploaded file content may be searched; Mask to AI controls whether recognised personal identifiers are replaced before AI processing.
Limits
- RebelVault is intended for documents, not bulk structured datasets. Use the data import flow for CSV or Excel data.
- Image-only PDFs may need OCR before upload.
- Very large documents should be split into smaller files.
- Search returns the most relevant chunks, not the full document.
- If a file is still embedding or embedding failed, it may not be available for answers.