Knowledge folders and files
Knowledge folders and knowledge files are how Atlan brings unstructured business content—SOPs, policies, and compliance rules—into the catalog as governed, discoverable assets. They give downstream AI agents access to procedural context that doesn't live in a database.
-
Curated, governed sources: Files are uploaded by the people who own the content, so what reaches downstream agents is a deliberate gold layer of authoritative documents rather than an automatic mirror of every drive or wiki.
-
Catalog-native assets: Folders and files appear in All assets with the same governance, lineage, and search affordances as any other Atlan asset.
-
Agent-ready procedural context: Context agents process uploaded files and extract glossary terms, business rules, and skills that flow into the Context Engineering Studio for downstream agent use.
Knowledge folder
A knowledge folder is a container asset that groups related documents together. A folder typically represents a domain or a policy area—for example, finance SOPs or compliance policies—and holds the files associated with that area.
Folders can be created by users or by AI agents. Agent-created folders are generated automatically as part of an agentic workflow and stay separate from user-created folders.
Knowledge file
A knowledge file is an individual document uploaded into a knowledge folder. Files are typically standard operating procedures, policies, or compliance rules that encode business logic in natural language rather than in a data model or schema.
Each file belongs to a single folder. The system blocks uploads with duplicate file names within a folder and retains the existing file. After upload, the file's profile page surfaces a preview and any extracted glossary terms and business rules that context agents produced from its contents.
AI agents can reference the same file across multiple folders as part of their workflows. User-uploaded files are always scoped to one folder.
How files reach context agents
Knowledge files and Context Repos in Context Engineering Studio form a one-way pipeline:
Knowledge files → context agents → skills, glossary terms, ontologies → CES Context Repo
| Knowledge files | Context Repos (CES) | |
|---|---|---|
| Layer | Upstream—raw organizational knowledge | Downstream—structured, agent-ready context |
| Content type | Unstructured: PDFs, SOPs, policy docs | Structured: skills, glossary terms, metric definitions, ontologies |
| What it encodes | Procedural context—how your org works with data | Structural context—what the data is and how it connects |
| Role in the pipeline | Source material uploaded by your team | Output of context agent processing for specific agent use-cases |
Unstructured sources in most organizations are fragmented across SharePoint, Confluence, and Google Drive, with multiple versions of the same policy and outdated SOPs in circulation. Rather than autoingesting all of that, knowledge files let the people who own the context choose the authoritative version—the goal is a curated gold layer, not a full mirror of every unstructured source.
A Knowledge Synthesis Agent crawls your unstructured sources, resolves conflicting document versions, deduplicates, and automatically proposes Knowledge Folders—making curation agent-assisted rather than fully manual, while keeping a human in the loop before content governs agent behavior.
Access control
Knowledge folder and file permissions are managed through bootstrap entity policies (pre-configured permission rules that Atlan seeds automatically when your tenant is created). The policies live in the Persona layer and use entity-type wildcards (entity:*), so they apply to all knowledge folders and files on the tenant.
| Action | Admin | Member | Guest | API token |
|---|---|---|---|---|
| View knowledge folders and files in the catalog | ✅ | ✅ | ✅ | ✅ |
| Preview and download file content | ✅ | ✅ | ✅ | ✅ |
| Create knowledge folders | ✅ | ❌ | ❌ | ✅ |
| Upload knowledge files | ✅ | ❌ | ❌ | ✅ |
| Edit or replace knowledge files | ✅ | ❌ | ❌ | ✅ |
Because the policies operate at the entity-type level rather than the asset level, two consequences are worth knowing:
- Members can't upload by default. Data stewards, owners, and team leads on the Member role can browse and preview knowledge content but can't contribute files without an admin performing the upload.
- No folder-level access control. Anyone who can see knowledge assets sees all knowledge folders and files on the tenant—you can't scope a persona to a subset of folders.
To extend upload access beyond admins, a tenant admin can create custom AuthPolicies that grant entity-create, entity-update, and entity-delete on KnowledgeFolder and KnowledgeFile assets for a dedicated persona. See Create a knowledge folder for steps.
Purpose policies (data product–style ACLs) aren't defined for knowledge assets yet. Custom AuthPolicies are the current supported method for extending upload access.
See also
- Extraction pipeline: How context agents process knowledge files and what they produce
- Create knowledge folder and upload files: Step-by-step instructions for uploading documents
- Working with Knowledge folders FAQ: File formats, size limits, and common questions
- Context Agents Studio: Configure and manage the context agents that process knowledge files