⚙️ Features

Locentra OS is a modular, self-learning LLM Operating System—engineered for crypto-native teams, LLM builders, and full-stack developers who need more than just a wrapper. It’s a complete runtime stack that gives you fine-tuning, memory, inference, and agent logic—all locally controlled.

---

🔁 Dynamic Fine-Tuning

Train your model on-the-fly—while it’s running.

Sources:

• Real-time prompt input

• Logged feedback from users

• High-frequency prompt patterns

• Autonomous scoring agents

No preprocessing. No restart. No external tools. Locentra triggers fine-tuning cycles dynamically based on usage and context.

Every response can become a training signal.

---

🧠 Semantic Memory Engine

Forget stateless chat history. Locentra builds contextual awareness over time.

Under the hood:

• Embedding-based vector recall using SentenceTransformers

• Top-k semantic similarity search across previous prompts

• Taggable, queryable memory logs

• Full persistence in PostgreSQL

Use it to:

• Build long-term memory

• Inject historical context

• Automatically adapt tone, intent, and domain

It doesn’t just respond—it remembers.

---

🤖 Autonomous Agents

Locentra includes an internal agent layer that watches, scores, and optimizes system behavior.

Agents can:

• Detect low-quality or overused prompts

• Trigger fine-tuning thresholds (AutoTrainer)

• Rewrite prompts and outputs (FeedbackLoopAgent)

• Improve clarity, correctness, and engagement dynamically

These agents are:

• Fully scriptable

• Chainable

• Deployable via CLI or config

• Built into the OS core

Your model evolves—even when you’re not watching.

---

🧰 Built for Developers

Locentra isn’t a black box. It’s built to be inspected, extended, and controlled.

Toolkit Includes:

• FastAPI backend

• React + Vite + Tailwind frontend

• Docker deployment stack

• PostgreSQL + SQLAlchemy database

• CLI tools for training, inference, and debugging

• Live logs, memory, and analytics dashboards

Everything is modular—every layer can be swapped, extended, or scripted.

Local-first, file-based, terminal-friendly.

---

🔐 Privacy-First Architecture

Locentra respects ownership—from tokens to training data.

• ✅ 100% self-hosted

• ✅ No API key needed

• ✅ No cloud calls

• ✅ No telemetry

• ✅ No vendor lock-in

• ✅ Full data transparency

Your data stays where it belongs: with you.