🛠 Under The Hood
From backend API layers to model routing and live vector memory, Locentra OS is built with a modern, container-native tech stack that’s made to scale—locally or in production. Everything is modular. Everything is swappable. Everything is under your control.
🧠 Backend Architecture
Language: Python 3.10+ Core Stack:
API Routing
FastAPI (async, OpenAPI-ready)
Database ORM
SQLAlchemy
Model Inference
Transformers (HuggingFace)
Embeddings & Vectors
SentenceTransformers
Web Serving
Uvicorn (dev) / Gunicorn (prod)
Built for LLMs. Tuned for async performance.
🖥 Frontend Stack
Everything you need for a fast, reactive developer UI—without the bloat.
Language: TypeScript Tools & Frameworks:
React– Component-driven architectureVite– Ultra-fast builds and hot module reloadTailwind CSS– Utility-first responsive design
Fast enough to feel native. Minimal enough to stay out of your way.
⚙️ DevOps & Infra
Locentra ships as a fully containerized application. No global Python installs. No backend/frontend mismatch. Just run and build.
Runtime
Docker, Docker Compose
Reverse Proxy
NGINX
Env Config
.env files (dotenv)
Scripting
Makefile, Bash utilities
Install in seconds. Launch in one command.
🧬 Model Compatibility
Whether you're working with lightweight test models or full-scale transformer architectures, Locentra adapts.
Tested Models (via HuggingFace):
Falcon(tiiuae/falcon-rw-1b, etc.)GPT-JMistralLLaMAAny other AutoModelForCausalLM-compatible transformer
Switch models by updating MODEL_NAME in .env.
All loading is dynamic. Hot reload supported.
🔄 Flexible, Swappable, Extendable
Locentra’s architecture isn't monolithic—it’s built in layers:
Swap vector engines
Replace tokenizer logic
Extend agents
Build custom adapters
Modify memory scoring
Plug in external tools
✅ Everything You Need, Nothing You Don’t
Locentra OS was built to feel like infrastructure—not magic. You can trace every call, customize every layer, and run it entirely offline.
Last updated
