Modern AI
Architecture.
A hybrid engine designed for universal AI continuity. Python for high-level reasoning. C++ for hardware-level performance. SQLite for ACID-compliant persistence.
Python Cognitive Brain
The Orchestration Substrate
RecallEngine
Intent-Driven routing → HNSW search → Hybrid Reranking (Cosine + Heuristics)
IntentDetector
Sub-10ms NLP Classifier across 6 core cognitive intents
ImportanceRanker
Neural-heuristic scoring: Goal (+3.0), Preference (+2.0), Feedback (+4.0)
ContextInference
Stateful analyzer mapping queries to optimal memory layer activations
C++ Moat
Hardware-accelerated engines for millisecond search at 1,000,000 nodes.
Verified search latency
Persistence
High-velocity ACID ingestion using SQLite WAL mode.
Model Interoperability
Standardizing context for every AI architecture
Recallix decouples memory from specific model parameters. Context stored by a reasoning model is fully accessible by a coding assistant, agentic framework, or research agent without semantic loss.
Atomic Store Pipeline
The 12-step propagation of a single memory.
Input
User interaction captured via POST /memory/store
Intent
Sub-10ms classification for routing
Sensory
Capturing raw buffer in 60s TTL deque
Working
Active context update & summarization
Episodic
SQLite WAL persistence with ID generation
Vector
128D Latent Embedding generation
HNSW
C++ HNSW index update (AVX2/SSE/NEON)
Linking
Unique ID mapping (Vector <-> SQL)
Ranking
Heuristic importance scoring
Gating
Long-term promotion threshold analysis
Reflection
Background LLM pattern extraction
Meta
Self-optimizing retrieval policy tune
Python 3.11
FastAPI, PyDantic, NumPy, Sentence-Transformers
C++ 17
HNSWlib, Eigen, Simd (AVX2/SSE/NEON), libcurl, nlohmann-json
SQLite 3
WAL Mode, Full-Text Search (FTS5), JSON1 extension
Next.js 14
Tailwind CSS, Framer Motion, Lucide, Recharts