Complicated underneath. Understandable above.
Heliarko works with ordinary things.
A photograph.
A PDF.
A letter.
A voice recording.
A project folder.
But storing a file is not enough if you want to find it by meaning, connect it with other material or ask useful questions about it later.
Heliarko is designed to build useful structure around the material in a Capsule.
You do not need to understand every technical detail to use Heliarko.
But you should be able to understand what the architecture is trying to do.
Three layers. That is the important part.
-
Your material
Photographs, documents, audio, video, notes and other things you choose to keep.
-
Knowledge structure
Heliarko creates structure around material so it can be searched, connected and used as context.
The original material remains the source.
-
Intelligence
Search and Synapse use available structure to find relevant context and help with a task.
Some functions may be local.
Some may use optional online AI.
Material → Structure → Intelligence
That is the basic architecture.
-
Keep the source.
The original document remains the source material.
An interpretation is not the original.
-
Read what can be read.
Where supported, text and useful information can become searchable.
-
Break large material into useful parts.
Long material may be represented in smaller sections for retrieval.
These sections are often called chunks.
-
Build a map of meaning.
Semantic representations such as embeddings and vectors can help find related material even when exact words differ.
-
Keep the way back.
The system should preserve a connection to the source wherever the implementation supports it.
Source first. Answer second.
You should not have to remember the filename.
Imagine that somewhere in your Capsule is a document about repairing the roof.
Unfortunately, the file is called scan_2024_08_17_final2.pdf
You probably will not remember that.
But you may remember: “the estimate for repairing the roof after the storm.”
Traditional filename search may fail.
Meaning-based search can look for material that is conceptually close to what you remember.
You remember the idea; Heliarko helps find the file.
You do not need this word to use Heliarko.
But there is no reason to make it mysterious.
An embedding is a numerical representation used to compare meaning.
For example: car repair and fixing the vehicle use different words but may describe related ideas.
An embedding can help place those meanings closer together.
The numbers themselves are not the memory.
They are part of a search structure.
A vector is simply a numerical form used in that comparison.
Large numbers of dimensions sound impressive.
They are not important to the person using Heliarko.
What matters is the result: better ways to find related material.
Technology should disappear when the explanation is finished.
A fluent answer is not enough.
AI is very good at producing sentences that sound convincing.
That does not automatically make those sentences true.
Where the current implementation supports source references, Heliarko should preserve the connection between useful context and its source.
The goal is not merely an AI that answers.
It is an AI that helps you check.
Context goes in. Help comes back.
A good system does not need every piece of a person’s digital life for every question.
Conceptually:
Your question
Search / retrieval
Relevant context
Synapse
Answer
Answer + sources, where source references are supported
This distinction is important.
The Capsule is built around local-first storage and structure.
That does not mean every possible form of AI must run on the computer.
Some intelligence may come from an external AI service when that function is used.
When that happens, the boundary should be visible.
Local Capsule
The durable personal foundation.
Online AI
An optional external intelligence layer.
What leaves the local boundary must be documented according to the real behaviour of the software.
Not guessed.
Not hidden.
Today’s AI will not be tomorrow’s AI.
AI models will change.
Providers will change.
Capabilities will improve.
Heliarko should not confuse one temporary intelligence provider with the permanent identity of the Capsule.
Your photographs are not an AI model.
Your family history is not an AI model.
Your documents are not an AI model.
Your context is not an AI model.
The intelligence used to work with them can evolve.
The durable part is the personal material and the structure around it.
The Capsule is the foundation. Intelligence is a tool connected to it.
AI can summarise a letter.
It does not replace the letter.
AI can describe a photograph.
It does not replace the photograph.
AI can infer a connection.
An inference does not become a historical fact simply because it sounds reasonable.
Heliarko should distinguish wherever practical between four different things.
Source
What actually exists in the material.
Context
Information connected to that source.
Inference
A conclusion derived from available information.
Generated content
Something created by AI.
These are different things.
A system built for long-term memory should not quietly mix them together.
You can safely ignore this section.
For people who want the terminology:
- Record
- A structured representation of an item or piece of information.
- Metadata
- Information about material — for example its type, date or descriptive properties.
- Chunk
- A smaller section of larger material used for search and context retrieval.
- Embedding
- A numerical representation used to compare meaning.
- Vector
- The numerical structure used in that comparison.
- Index
- An organised structure that helps find relevant material efficiently.
- Context
- The information selected as relevant to a particular question or task.
- Reference
- A connection back to source material.
- Model
- The AI system used to process a request or generate a response.
You do not need to learn any of these words.
Heliarko does.
Storage is not memory.
A disk can store a million files.
Search is not understanding.
Finding a word is not the same as finding the idea.
AI is not truth.
A confident answer still needs checking.
Context is the bridge.
It connects stored material with useful intelligence.
That bridge is one of the central technical ideas behind Heliarko.
Storage formats will change.
Search technology will change.
AI models will change.
Hardware will change.
Heliarko should be able to improve those layers without making a person’s life dependent on one fashionable technology.
This is why the architecture begins with the Capsule rather than the AI provider.
The human material is the durable asset.
Everything around it is replaceable technology.
- Product type
- Local-first personal digital environment
- Primary personal container
- User-controlled Capsule
- Primary material
- Documents, photographs, audio, video, notes and personal records
- Knowledge structure
- Structured search/context data according to the current implementation
- AI layer
- Synapse
- Online AI
- Optional where supported; exact network behaviour requires technical verification
- Primary principle
- The Capsule remains the foundation. AI remains an intelligence layer.
- Current official technical source
- heliarko.com/technology/
Technology should earn trust by becoming understandable.
You do not need to know what a vector is to preserve a photograph.
You do not need to understand embeddings to find an old document.
And you should not need to be an engineer to understand where your own information goes.
That is the standard.