From Business Records to Organizational Knowledge
This chapter examines how Ever Gauzy extends its shared operational context from structured business records into documents, search, and AI-assisted work. The focus is not whether a model can produce a fluent answer. It is how a document moves from upload or capture through extraction, review, indexing, and retrieval, and how tenant credentials and user permissions constrain what an embedded assistant can do. That mechanism lets readers judge both the practical value of Gauzy’s knowledge layer and the fallbacks required when AI, queues, or vector search are unavailable.
Documents become useful when they stay connected
A file store answers where a file is. A knowledge layer also needs to answer what the content means, who may use it, which business record it belongs to, and whether the content has been reviewed. Gauzy’s Documents plugin is designed as a hub for uploaded files, authored pages, folders, categories, versions, links, and shares. Its central document entity distinguishes folders, pages, and files, while supporting records hold versions, links to other business entities, sharing state, extracted chunks, index state, and inbound capture addresses.
That structure gives a document a place in the same organization context as a project, client, employee, or task. A project brief can be linked to a project. A policy page can belong to an organization’s knowledge tree. An uploaded file can be associated with a business record rather than left under an unsearchable personal folder. A version preserves the fact that authored content changes over time. A share record and permission gate help distinguish “the organization has this document” from “this user may read it.”
The distinction is important because documents often contain more context than their filename suggests. A proposal may contain pricing and client commitments. An employee document may contain sensitive personal information. A task attachment may include technical or contractual detail. The Documents routes are guarded by tenant and permission checks and by the Documents feature flag. Those gates make the knowledge layer part of Gauzy’s existing authorization model rather than a side channel around it.
Inbound email shows why provenance and review matter. Gauzy can receive documents through an organization capture address, but the plugin describes those arrivals as pending and keeps them out of the AI knowledge base until a human approves them. A custom-domain address must prove domain ownership, and signed delivery proofs are checked before the recipient is resolved. OCR-derived text is also treated as reviewable because transcription can lose or distort source content. These controls recognize that content entering a knowledge base is not automatically trustworthy merely because it arrived through a supported route.
The same principle applies to authored pages and uploaded files. A versioned page may be authoritative for an organization’s current process, while an old upload may be useful historical evidence but not current instruction. A document link provides business context, not a guarantee that the document is accurate. Search makes content easier to find; it does not decide which content should govern a decision. Those distinctions must remain visible when AI is added later.
The link between a document and a business record also changes how access should be reasoned about. A project document may be visible to project members but not to every employee in the organization. A client-facing share may expose a deliberately limited view without granting access to internal versions or extracted chunks. An employee document may need a narrower audience than a public project page.