Composable layers — each one a clearly bounded transformation logic.
AI frameworks compose like layers in a matrix — not as a literal linear-algebra operation, but as an intuition for how independent layers combine to produce richer outcomes. Rather than building one monolithic framework that tries to cover everything, the architecture follows a principle of modular composition.
Domain-agnostic by design. Defines fundamental guardrails and the 3-step lifecycle — how AI-assisted development proceeds, not what is built.
Specialized vocabulary, constraints, patterns, quality criteria for a field (games, UI, enterprise). Composed with the base via context injection.
Real-time or proprietary data injected at query time. Higher token cost; richer context. Does not modify static layers.
Domains stay independent modules instead of forming one large inseparable system.
Domains can be added, replaced, or evolved in isolation without affecting the core.
Teams work on different domain layers concurrently, all anchored to the same stable base.
Only the domains relevant to a given project are injected — context stays lean.
| Layer | Role | Stability |
|---|---|---|
| Core Framework | Process, lifecycle, guardrails | High — rarely changes |
| Domain Frameworks | Domain-specific patterns & constraints | Medium — evolves with the domain |
| External Knowledge | Real-time / proprietary context | Dynamic — changes continuously |