System Overview
Architecture and core principles of the Cortyxia Memory Management Unit.
Architecture Overview
Cortyxia implements a distributed Memory Management Unit (MMU) that provides intelligent context orchestration between applications and LLM providers. The system operates as a middleware layer that abstracts memory operations, enabling semantic retrieval, token optimization, and cross-provider consistency.
The architecture follows a multi-tier design with distinct separation of concerns: SDK interface layer, core proxy engine, memory storage subsystem, and observability pipeline. This modular approach enables horizontal scaling, fault isolation, and independent component evolution.
Core Components
Proxy Engine
High-performance Rust-based proxy (iso-core) that handles request routing, token encoding/decoding, and context window management. Built with async I/O for sub-200ms latency end-to-end.
Memory Storage Layer
Content-addressable storage (CAS) with SHA-256 hashing for deduplication and integrity verification. The core proxy uses SQLite by default. PostgreSQL powers telemetry and analytics, and Redis is optional for caching.
Query Engine
Hybrid retrieval system combining BM25 indexing with semantic reranking. Built on Tantivy search engine for inverted index operations with configurable TF-IDF weighting and cross-encoder precision scoring.
Security Layer
Project-scoped namespace isolation with cryptographically secure token encoding. Implements role-based access control (RBAC), audit logging, and encryption-at-rest for enterprise compliance requirements.
Observability Pipeline
Real-time telemetry collection with metrics for token efficiency, latency percentiles, memory hit rates, and knowledge debt analysis. Supports Prometheus export at /metrics and is available through the OSuite dashboard and the SDK observability namespace.
Performance Characteristics
Deployment Architecture
Cloud-Hosted
Fully managed service where Cortyxia hosts all infrastructure and code. Your data resides in our cloud with automatic scaling, 99.9% uptime SLA, and zero infrastructure maintenance overhead.
Self-Hosted
Run Cortyxia on your own infrastructure. The core proxy uses SQLite by default, with PostgreSQL for telemetry and analytics and optional Redis for caching. You keep full data sovereignty.