For years, I organised my personal archive like many engineers probably do.
I bought a NAS.
I created folders.
I invented naming conventions.
I was convinced that if the structure was clean enough, future me would thank present me.
Future me was… moderately grateful.
The system worked.
Until it didn’t.
Documents/
├── Finance/
│ └── Taxes/
│ └── 2025/
│ └── declaration.pdf
│
├── Home/
│ └── Electricity/
│ └── invoice.pdf
│
└── Work/
└── Contracts/
└── employment.pdfThe problem was not storage.
Storage was easy.
Modern document systems are excellent at ingesting files, extracting text, running OCR, indexing content, and making documents searchable.
The problem was something else:
A document rarely belongs to only one place.
The problem with folders
A folder forces every document to have a single home.
But reality does not work like that.
A rental contract is not only a legal document.
It is also related to:
- a property
- a tenant
- a landlord
- a country
- a period of time
- financial transactions
An electricity invoice is not only a bill.
It is evidence of:
- a service provider
- a house
- a contract
- a payment
- a consumption period
The information exists.
The documents exist.
But the relationships are hidden.
Usually somewhere between my memory, a spreadsheet, and a folder called Misc that nobody wants to open.
A folder captures location.
It does not capture meaning.
From hierarchy to relationships
At some point, I stopped asking:
Where should I put this document?
and started asking:
What is this document about?
That small change completely changes the model.
A document is not just a file.
It is evidence about things that exist in the real world.
The invoice is no longer something sitting inside a folder.
It becomes a piece of evidence connected to a larger knowledge system.
The same problem I had seen before
The interesting part is that this was not really a new problem.
During my engineering work, I repeatedly encountered the same pattern:
Information is everywhere.
Meaning is not.
Data exists in databases, documents, reports, emails, and people’s heads.
The difficult part is not storing more information.
The difficult part is preserving the context that makes information useful.
AtlasDocs started as a personal archive experiment, but the underlying question is much broader:
How do we build systems where information remains understandable over time?
Why not just extend Paperless-ngx?
When I started exploring this idea, Paperless-ngx was the obvious foundation.
It already solves many difficult problems:
- document ingestion
- OCR
- search
- previews
- permissions
- lifecycle management
The mistake would be asking it to solve a different problem.
Paperless is excellent at answering:
Where is my document?
But I needed something answering:
What does this document mean, and how does it connect to everything else?
Trying to embed a semantic model directly into Paperless would couple two different responsibilities.
Instead, AtlasDocs adds a semantic layer on top of it.
Paperless remains the system of record for documents.
AtlasDocs understands what those documents represent.
The AtlasDocs model
The core idea is simple:
A document is an entity that can participate in relationships.
Instead of:
invoice.pdfwe model:
The goal is not to create a giant graph database.
The goal is to add the missing semantic layer that allows information to remain understandable as the archive grows.
Design principles
AtlasDocs follows a few principles.
Keep the existing archive
Documents remain where they belong.
AtlasDocs does not replace Paperless.
It references it.
AtlasDocs UUIDExternalReferencePaperless Document ID
Prefer explicit relationships over hidden assumptions
A tag says:
TaxA relationship says:
Meaning becomes inspectable.
Preserve provenance
Knowledge without origin becomes unreliable.
Every relationship should eventually answer:
- Where did this information come from?
- Was it manually created?
- Was it imported?
- Was it inferred?
Current state
AtlasDocs is still early.
The current version focuses on the foundation:
- Entity model
- External references
- Typed relationships
- Ontologies
- Reconciliation with Paperless
- Classification workbench
Not yet:
- LLM enrichment
- embeddings
- automatic extraction
- graph visualisation
- intelligent assistants
Those are possible because the semantic foundation exists first.
Why this matters beyond documents
The interesting part is not document management.
Documents are only one example.
The same problem appears everywhere:
- engineering knowledge
- company processes
- research
- compliance
- personal archives
Information is usually abundant.
Meaning is the scarce resource.
AtlasDocs is an experiment around one question:
What happens when we stop organising information around files and start organising it around knowledge?
AtlasDocs — an open source knowledge architecture experiment built on top of Paperless-ngx.