Upload knowledge documents

You have a glossary spreadsheet, a vendor’s data dictionary, and a requirements document from the business. You want DeltaVault’s AI to draw on all three when it recommends models, drafts descriptions, or answers questions. The knowledge library is where they go. The one decision that matters at upload time is the document type, because the type controls how the AI reads the document later:

• Glossary and Business requirements documents are given to the AI in full. Every term and every requirement is present whenever the AI works.
• Domain knowledge documents are searched in pieces. The AI retrieves the sections relevant to the task at hand.

File a requirements document under domain knowledge and it still uploads fine, but the AI now sees fragments of it instead of the whole thing, and nothing warns you that requirements are missing. Pick the type that matches what the document is, not where it came from.

The three types

The upload dialog describes each type on its card:

Type	Card description
Glossary	”Term, acronym, and code definitions; business data dictionaries.”
Business requirements	”Use cases, KPIs, reporting needs, data quality rules.”
Domain knowledge	”Industry context, business rules, source system documentation, modeling standards, regulatory notes.”

A vendor’s data dictionary or a source-to-target mapping goes under Domain knowledge, which is where source system documentation now belongs.

Organization or project scope

Every document has a scope, and the scope is always your call:

• Organization: the document informs AI work across your whole organization.
• Project: the document informs AI work only within the project you pick.

Let AI decide picks the type only; it never moves a document between organization and project scope. When you upload from inside a project, the scope is locked to that project and the dialog says so.

After upload, the documents grid on the Knowledge Base page groups documents by their effective location. Organization-scoped documents appear under an Organization heading; project-scoped documents appear under a heading named after the project they belong to. Each heading shows the document count for that group. There are no separate Scope or Project columns: the group heading tells you where a document lives.

The grid fills the available screen height and scrolls internally. Groups are expanded by default: click a group heading to collapse or expand it.

Upload a document

1. Go to Knowledge and click Upload.
2. Under 1. Type, pick a type card (or pick Let AI decide if you are not sure).
3. Under 2. Scope, choose Organization or Project. For project scope, pick the project.
4. Under 3. File, attach the file. Accepted formats: .pdf, .txt, .md, .docx, and .csv.
5. Add a description or comment if useful: both are optional.
6. Click Upload.

The document appears in the grid under its location group (Organization or the project name) with status Processing while DeltaVault indexes it, then flips to Ready.

Drag and drop a file

Instead of clicking Upload, you can drag a file from your computer straight onto the document grid (whether it is empty or already has rows). DeltaVault opens the upload dialog with the file already attached, so you only need to pick the document type and scope, then confirm.

One file at a time. Supported file types are PDF, TXT, MD, DOCX, and CSV; other types are declined with a notice.

What “Let AI decide” does

The card says it plainly: “Not sure, or mixed content? DeltaVault reads the file after upload and picks the type. Mixed files become a split proposal for your review.” Here is what that looks like:

1. After upload, the Type column shows Classifying… while the AI reads the file.
2. The AI picks a type, and the document continues through normal processing.
3. The chosen type appears in the grid with a small sparkle icon next to it. Hovering over the icon shows Type chosen by AI.
4. The document’s page carries the same attribution with a confidence figure, for example Chosen by AI · 85% confidence.

The AI’s choice is yours to overrule. Open the document, click Edit, pick a different type, and save: the document reprocesses under your type and the AI attribution disappears. Editing anything else, such as the name or description, keeps the attribution.

Mixed files

A file that genuinely mixes content kinds (a vendor manual that is half data dictionary and half business rules) would be short-changed by any single type. When the AI detects this, it does not guess. It proposes splitting the document into parts, one per type, and parks the original until you review the proposal. See Review a split proposal.

When classification fails

Could not classify: set a type manually. The AI could not settle on a type for this file; the underlying reason is appended to the message in parentheses. The document sits in an error state and stays out of search and AI context. Open the document, click Edit, pick a type, and save. Processing restarts under your chosen type, and the classifier stays out of the way.

Mixed content detected: consider splitting manually. The AI judged the document mixed, but the file is too large to split faithfully, so it kept the document whole under the dominant type and left this note on the document’s page. Nothing is blocked: the document is processed and searchable under that single type. If the mix matters, split the file yourself and upload the parts as separate documents.

Editing and checking documents with AI

Once a document is in your library, AI can help you keep it sharp and aligned:

• Editing knowledge with AI rewrites a passage or a whole document, and you accept or reject each change before saving.
• Governance check cross-references a document against your business entities and category lanes and shows you the coverage and the gaps.