Architecture

The SoulCore V4 ΔState Architecture

The foundation of SoulCore is the V4 ΔState model. The operation between language and state is not a fixed vector, but a state transition: the system computes a ΔState from the current state and the input, and moves to a new state through it.

The full pipeline

User → Language Runtime Encoder (300M) → 768-dimensional state → Hybrid Core V8 (200–300M) → ΔState → Language Runtime Decoder (300M) → Text → Response

Cognitive Process Steps

Thinking occurs in eight steps:

1. Information — ingoing user text
2. Processing — Language Runtime Encoder (300M) → CSF representation
3. Understanding — CSF → intent, goal, priority
4. Intent — ChangeHead: „Is it worth acting?”
5. Reflection — Predictor: „What will happen?”
6. Ideation — Planner: „What should I do?” (alternatives)
7. Decision — Verifier: „Is it safe?” + Explorer: „Is there another way?”
8. Response — Language Runtime Decoder (300M) → Text

StateBundle

The StateBundle is a 768-dimensional structured state describing the current condition of the system. It is not a homogeneous vector, but a bundle of five explicit segments — so each core knows exactly where to read from and write to.

• Internal state (128) — emotion, focus, internal stability
• World state (256) — external facts, knowledge base
• Goal state (128) — short and long term intents
• Memory state (128) — context carrier
• Execution state (128) — currently running actions

Total: 128 + 256 + 128 + 128 + 128 = 768 dimensions.

Hardware Requirements

SoulCore development uses two NVIDIA RTX 5060 Ti graphics cards, totaling 32 GB VRAM.

Architectural Components

Language Runtime (300M encoder + 300M decoder)

Responsible for text ↔ state translation. This model does not think — it only translates. The encoder produces a 768-dimensional representation from language, and the decoder turns the selected state back into text.

Hybrid Core V8 (200–300M)

The brain that thinks in states. MLP network for short-term processing, RNN network for long-term memory.

Runtime Kernel

The infrastructure that manages state, memory, and goals. This is not a model — this is an operating system.