yecoai-cognitive-layer 0.1.0

Lightweight cognitive protection layer for LLM systems

🧠 YecoAI Mini-LLM Cognitive Layer

Lightweight cognitive protection for Large Language Models

Anti-loop • Amnesia detection • Semantic stability

Developed by www.yecoai.com

---

✨ What is this?

YecoAI Mini-LLM Cognitive Layer is a lightweight, modular guard layer designed to sit on top of any LLM.

It does not replace the model.
It observes, evaluates, and stabilizes the model’s output in real time.

Built to solve real production problems:

• Infinite loops
• Context loss (amnesia)
• Semantic collapse in long conversations
• Unstable autonomous agents

---

🧩 Core Capabilities

• 🔁 Loop Detection
  Identifies structural and semantic repetition patterns.

• 🧠 Amnesia Detection
  Detects loss of contextual continuity across turns.

• 🧯 Semantic Degradation Guard
  Protects against meaning collapse over time.

• ⚡ Ultra-Low Resource Usage
  Designed for embedded systems and edge deployments.

---

📊 Benchmark Results (v1.0)

Real stress tests. No synthetic demos.

Test Suite

• 142 extreme stress scenarios
• Multilingual semantic traps
• Long-context degradation
• Loop-inducing prompts

Total Accuracy:         76.06%
Loop Detection (F1):    0.90
Normal Detection (F1):  0.71
Amnesia Detection (F1): 0.63
Average RAM Usage:     38.85 MB

✅ Loop detection is currently the strongest and near production-ready.
⚠️ Amnesia detection is functional but still evolving.

Detailed reports are available in /benchmarks.

---

🏗️ High-Level Architecture

LLM Output
↓
Cognitive Evaluation Layer
├── Loop Detector
├── Amnesia Detector
└── Semantic Stability Guard
↓
Validated / Flagged Output

---

🚀 Use Cases

• Autonomous AI agents
• Long-running chat systems
• AI copilots & assistants
• Embedded / edge AI
• Guard layers for SaaS AI products
• LLM research & experimentation

---

🧪 Project Status

• Version: v3.0 (Stress-Tested Edition)
• Maturity: Experimental / Research-grade
• Focus: Stability, efficiency, interpretability

This repository is part of the YecoAI Cognitive Systems stack.

---

🏷️ Attribution & Credits (Required)

This project is developed and maintained by YecoAI.

Attribution is REQUIRED in any usage, including:

• Modified versions
• Commercial products
• SaaS platforms
• Research publications
• Closed-source integrations

You must retain:

• This README attribution
• The LICENSE file
• The NOTICE file

---

📄 License

Licensed under the Apache License 2.0.

✔ Commercial use
✔ Modification
✔ Redistribution
✔ Closed-source integration

Attribution and preservation of notices are mandatory.

See LICENSE and NOTICE for details.

---

🌐 About YecoAI

YecoAI builds next-generation cognitive systems focused on:

• AI stability & safety
• Autonomous agents
• Real-world deployability
• Low-overhead intelligent layers

Website : https://www.yecoai.com Discord : https://discord.gg/rBZscZtMvX GitHub : https://github.com/YecoAI

---

© 2026 www.yecoai.com
Original author: Marco (HighMark / YecoAI)