01APPLICATIONS
Application Layer
Enterprise systems, Spring Boot microservices, APIs, and product experiences consuming deterministic AI capabilities.
Spring Boot 3/4MicroservicesREST EndpointsDomain Logic
↗
←Back to Shree AI OS
SHREE AI OS · ARCHITECTURE
01The architecture
The architecture
behind
intelligent software.
Shree AI OS is structured as a layered platform where applications, runtime services, intelligence capabilities, and foundational infrastructure work together through explicit boundaries.
v1.0.6 · ARCHITECTURE FOUNDATION
02 · SYSTEM OVERVIEW
One platform.
Defined layers.
The architecture separates applications from the underlying intelligence and infrastructure while keeping the complete system connected through explicit interfaces.
01
APPLICATIONSApplication Layer
02
INTERFACESSDK Layer (10 Verified Facades)
03
RUNTIMERuntime Orchestration Layer
04
KERNELSKernel Services Layer
05
PROVIDERSLLM Provider Layer (BYOK)
03 · ARCHITECTURE LAYERSEvery layer
Every layer
has a responsibility.
The architecture is intentionally layered so that applications can evolve independently while the platform continues to provide reusable infrastructure and intelligence.
02INTERFACES
SDK Layer (10 Verified Facades)
Public Java 21 client interfaces providing typed, deterministic access to memory, planning, code patches, and model routing.
client.chat()client.memory()client.knowledge()client.planning()client.developer()
↗
03RUNTIME
Runtime Orchestration Layer
Deterministic 11-stage cognitive execution pipeline, fail-closed authorization gate, multi-tenant boundaries, and event dispatching.
11-Stage PipelineFail-Closed GateIntent RouterTenant Enforcer
↗
04KERNELS
Kernel Services Layer
In-process cognitive kernels for episodic memory, pgvector hybrid RRF search, topological DAG planning, and AST code patch generation.
Episodic Memorypgvector HNSW + GINK0.6 AcquisitionJavaParser AST
↗
05PROVIDERS
LLM Provider Layer (BYOK)
Swappable model integration with automatic HTTP 429/503 exponential backoff retries and deterministic in-memory fallback.
Google GeminiOpenAIOllamaIn-Memory Fallback
↗
04 · INTELLIGENCE FLOW
From intent
to execution.
Intelligence becomes useful when it can move through a controlled path from application intent to context, intelligence capabilities, and execution.
01Identity & ContextActor Resolution & Tenant Boundaries
↓
02Memory & KnowledgeDual-Channel RRF & K0.6 Acquisition
↓
03Reasoning & InferenceFact Checking & Hypothesis Scoring
↓
04Planning & ExecutionTopological DAG & Fail-Closed Gate
↓
05Reflection & ReviewAdaptive Calibration & Governance
↓
06Dual-Mode SynthesisStrict RAG Citations vs. General Fallback
05 · DESIGN PRINCIPLESArchitecture
Architecture
before complexity.
The architecture is guided by principles intended to keep intelligence infrastructure understandable, reusable, and capable of evolving over time.
Separation
Applications remain separated from internal platform implementations through explicit architectural boundaries.
Reusability
Intelligence capabilities are designed as reusable foundations rather than application-specific implementations.
Modularity
Each architectural layer has a defined responsibility and can evolve without unnecessarily coupling the entire system.
Extensibility
The architecture provides room for new capabilities, applications, and intelligence systems as the platform evolves.
ARCHITECTURE → PLATFORMArchitecture gives
Architecture gives
the platform structure.
The architecture defines how the platform is organized. The platform turns those architectural boundaries into reusable infrastructure for applications and developers.
Explore Platform↗