RCT OS Systems Surface
10-layer cognitive architecture for AI systems that must stay provable.
This architecture page now acts as a systems surface that bridges the RCT OS CLI language with the platform's real footprint, keeping public SDK proof, enterprise runtime footprint, and benchmark scope visibly separate.
RCT OS
Architecture Rail
Why 10 Layers?
Inspired by the OSI networking model but designed specifically for AI systems. Each layer has a clear responsibility boundary, enabling independent scaling, testing, and evolution.
Unlike monolithic AI frameworks, the layered approach allows enterprises to adopt specific layers incrementally. Start with L6 for orchestration, add L8 for safety, then expand as needs grow.
The architecture currently spans a 52+ runtime-component footprint across all 10 layers, with scaling designed to preserve governance, observability, and operational clarity.
Explore Each Layer Interactively
Select a layer to inspect the service groups and platform capabilities that make up the 10-layer RCT architecture.
All 10 Layers at a Glance
Hardware Abstraction
GPU/TPU management, resource allocation, and infrastructure orchestration across cloud and edge deployments.
Data Ingestion
Multi-modal data pipeline supporting text, images, audio, video, and structured datasets with real-time streaming.
Knowledge Engine
Semantic indexing, vector search, and knowledge graph construction for intelligent information retrieval.
Memory & Context
RCTDB v2.0 with the 8-dimensional universal memory schema — Identity, Sovereignty, Context, Payload, Value, Social, Delta, Verification — for persistent context across sessions and agents.
Reasoning Core
Multi-strategy reasoning including chain-of-thought, tree-of-thought, and hybrid approaches powered by the FDIA Equation.
Multi-LLM Orchestration
The model is a setting: a local model, any model on OpenRouter, or any OpenAI-compatible endpoint, with a fast or slow path per task and backup models. The planned seven-model roster does not run.
Agent Framework
Autonomous agent lifecycle management with JITNA Protocol for inter-agent communication and consensus.
Safety & Verification
A safety gate before tool calls, human-signed approvals, a signed audit log and an optional jury of independent models. No hallucination reduction has been measured.
Application Layer
Domain-specific assistants, enterprise workflows, and solution packages built on the shared platform core.
Self-Evolving Orchestrator
Continuous self-improvement through performance monitoring, A/B testing, and adaptive algorithm selection.
How RCT Compares
| Feature | RCT Ecosystem | Others |
|---|---|---|
| Architecture | 10-Layer Cognitive Stack | Monolithic / 2-3 layers |
| Memory | RCTDB v2.0 — 8D Schema | No persistent memory |
| Hallucination | Not claimed (no measurement) | Not compared |
| Model choice | A setting: local, OpenRouter or any OpenAI-compatible endpoint | Often one provider |
| Self-Improvement | Skills learned from verified episodes; new tools only with a human signature | Manual updates only |
| Protocol | JITNA open standard | Proprietary APIs |
Explore More
Evidence Lanes Snapshot
Read this architecture page with the public SDK proof lane separated from the enterprise runtime footprint and benchmark scope.
Ready to Build on the 10-Layer Stack?
Frequently Asked Questions
What is the Delentia Labs 10-Layer AI Architecture?
The Delentia Labs 10-Layer Architecture is a constitutional AI stack spanning from hardware abstraction at layer 1 to self-evolving orchestration at layer 10. Each layer serves a distinct role in ensuring verifiable, auditable, and governed AI execution across enterprise deployments.
How does the optional jury work in the architecture?
The jury is optional. Several models you name each vote agree, disagree or abstain independently, and the verdict is signed. It has been tested with scripted voters only; it has not been shown to remove single-model bias or to reduce hallucination.
What is the role of constitutional AI in RCT's architecture?
Constitutional AI principles are embedded at every layer of the RCT stack. They define what the AI system can and cannot do, how conflicts are resolved, and how auditability is maintained. This ensures AI decisions remain aligned with organizational policies and ethical constraints.
How many runtime components does the architecture include?
The production architecture includes 52+ runtime components distributed across 10 layers, covering memory, routing, verification, orchestration, and self-evolution. Each component is independently deployable and composable.
Can the architecture be deployed on-premise?
Yes. The Delentia Labs architecture supports on-premise, hybrid, and cloud deployment models. Enterprise clients can isolate specific layers for data sovereignty compliance, particularly in regulated markets such as Thailand under PDPA.