1 · Motivation
The end of human-written code changes the work.
For decades, the limiting factor in software engineering was the pace at which skilled human developers could write and review code. That limit no longer holds. AI coding systems can generate well-structured code across major programming languages in seconds, and the transition is not merely a supplement to human coding.
This has a dramatic consequence: the volume of code in the world will explode. Systems that once required years of coordinated engineering can now be produced and transformed at a velocity no prior organization could have achieved.
That layer cannot be code. Code is what the AI produces, not what humans meaningfully reason about. The shared layer must sit above code: at the level of architecture, meaning, and intention.
2 · Existing Systems
Legacy software must be understood before it can be changed.
The software that already runs the world exists now, in millions of lines of code accumulated over decades. The teams that originally built it have dispersed, documentation is outdated, and implicit knowledge has been forgotten.
Before any AI system can evolve, refactor, or extend a large legacy system, it must understand what that system already does in precise, structured, navigable detail. That understanding cannot come from outdated documents or one-file-at-a-time analysis.
Siloquy therefore aims for a persistent, machine-readable representation of meaning: a model that says not just what lines of code are present, but what each part of the system does and why it exists.
3 · Approach
The higher-level model must remain grounded in implementation reality.
Programming languages already provide the meta-language of software: functions, conditions, loops, calls, modules, classes, interfaces, and containment. Any higher-level representation must take that grammar seriously.
Siloquy builds the higher-level model as a structured superset of what can be implemented. Every abstract concept maps to something real in the code below it. Every relationship remains traceable. The higher level adds structure and meaning; it does not replace implementation reality with fiction.
UML, extended where necessary, supplies a proven structural vocabulary. AI removes the old constraint that diagrams and models were too expensive to create and keep synchronized by hand.
4 · What Siloquy Is
A layered semantic digital twin of software.
Siloquy transforms source code into a structured, navigable, machine-interpretable representation derived directly from implementation. It connects four levels of abstraction:
These layers are explicitly linked. A reader can move from high-level intention down to code lines, or from any fragment of code back up to its role in the overall system.
5 · Design Outline
Five sequential layers turn code into navigable meaning.
Builds a complete language-neutral structural graph: SCIR, organized by lexical flow, unit connectivity, data and conditions, async execution, and containment.
Reduces structural artifacts into MeaningCells, then forms MDG for dependency order and TOM for logical execution paths.
Annotates each MeaningCell bottom-up with what that part of the system does, grounded in source evidence.
Annotates each MeaningCell top-down with why it exists, including external units, boundary calls, and implied entities.
Defines graph selection, projection, abstraction, and expansion so humans and agents navigate the same model.
6 · Access Model
One model for visual exploration and agentic traversal.
The Siloquy Access Model is not a user interface. It is the formal projection and navigation model for the knowledge produced by Layers 1 through 4. It exposes structural, execution, semantic, and boundary views from the same underlying graph.
Its design principle is simple: humans and AI systems collaborating on software must share not only the knowledge base, but also the conceptual framework for navigating it.