---
title: Codette Demo
emoji: 🧠
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.9.0
python_version: '3.10'
app_file: app.py
hf_oauth: true
pinned: false
---

# Codette Demo - Multi-Perspective AI Reasoning

Interactive demo of the **Codette Multi-Perspective Reasoning System** - an AI that approaches problems through 9 specialized cognitive lenses simultaneously.

## What is Codette?

Codette is not just another chatbot. It is a reasoning architecture that:

- **Debates internally** across 9 specialized perspectives before answering
- **Measures epistemic tension** between viewpoints (semantic tension engine)
- **Monitors reasoning health** in real-time (coherence field / Gamma metric)
- **Routes intelligently** by query complexity (SIMPLE/MEDIUM/COMPLEX)
- **Validates ethically** through 3-layer ethical governance (EthicalAIGovernance)
- **Remembers** reasoning exchanges in persistent cocoon memories (CognitionCocooner)
- **Self-corrects** constraint violations before sending responses
- **Obeys 4 permanent behavioral locks** baked into all 9 adapters through training
- **Substrate-aware** -- adjusts reasoning based on real-time hardware pressure
- **Self-introspects** -- analyzes her own cocoon history for real measured patterns
- **AEGIS ethics** -- 6-framework ethical evaluation on every response

## The 9 Perspective Adapters

| Adapter | Cognitive Lens | Specialty |
|---|---|---|
| Newton | Analytical | Physics, systematic reasoning, empirical evidence |
| DaVinci | Creative | Cross-domain invention, visual thinking, design |
| Empathy | Emotional | Human experience, feelings, relationships |
| Philosophy | Conceptual | Ethics, fundamental questions, logic |
| Quantum | Probabilistic | Uncertainty, superposition, complementarity |
| Consciousness | Recursive | Meta-cognition, RC+xi framework, self-reflection |
| Multi-Perspective | Integrative | Cross-lens synthesis, holistic understanding |
| Systems Architecture | Engineering | Modularity, scalability, design patterns |
| Orchestrator | Coordination | Query routing, debate management, coherence |

## Architecture (Phase 6+)

```
Query Input
    |
    v
[Executive Controller] -- Classifies: SIMPLE / MEDIUM / COMPLEX
    |
    v
[Substrate Monitor] -- Adjusts routing based on system pressure
    |
    v
[Adapter Router] -- Selects optimal perspective(s)
    |
    v
[Code7eCQURE] -- Emotional context enrichment (quantum cocoon)
    |
    v
[Multi-Agent Debate] -- LLM inference with semantic tension tracking
    |
    v
[Coherence Field] -- Gamma monitoring, FFT collapse detection
    |
    v
[Colleen Conscience] -- Emotional + ethical validation
    |
    v
[AEGIS Ethics] -- 6-framework evaluation (eta alignment score)
    |
    v
[Guardian Spindle] -- Safety + trust calibration
    |
    v
[Synthesis] -- Unified multi-perspective response + cocoon storage
```

### Key Framework Components

- **Semantic Tension Engine (xi)**: Quantifies disagreement between perspectives
- **Coherence Field (Gamma)**: Real-time metric detecting reasoning collapse
- **Quantum Spiderweb**: Belief propagation across adapter perspectives
- **AEGIS Governance**: 6-framework ethical validation (utilitarian, deontological, virtue, care, ubuntu, indigenous reciprocity)
- **Memory Kernel**: Emotional continuity via SHA256-anchored cocoon memories
- **Cocoon Stability Field**: FFT-based detection of repetition and vocabulary collapse
- **Substrate Awareness**: Hardware-aware cognition -- adjusts under pressure like biological fatigue
- **Cocoon Introspection**: Self-analysis of reasoning history (adapter dominance, emotional trends, pressure correlations)
- **Code7eCQURE**: Quantum emotional context enrichment on every query

## How It Works

1. **You ask a question** - anything from simple facts to deep philosophical puzzles
2. **Executive Controller** classifies complexity and routes accordingly
3. **SIMPLE queries**: Direct answer (~150ms, skips heavy machinery)
4. **MEDIUM queries**: 1-round debate with 2 perspectives (~900ms)
5. **COMPLEX queries**: Full 3-round debate with all perspectives (~2500ms)
6. **Semantic tension** tracked between perspectives throughout
7. **Coherence field** monitors for reasoning collapse
8. **Final synthesis** integrates insights with ethical validation

## Try It

Example prompts that showcase multi-perspective reasoning:

- "What would it mean for a machine to genuinely understand something?"
- "Is mathematics discovered or invented?"
- "How should we balance individual privacy with collective security?"
- "Explain consciousness from multiple perspectives"
- "Design a system that learns from its own mistakes"

## Technical Details

| Component | Details |
|---|---|
| Base Model | Llama 3.1 8B Instruct |
| Quantization | Q4_K_M (GGUF, ~4.6 GB) |
| Adapters | 9 LoRA adapters (~27 MB each, GGUF format) |
| Inference | llama.cpp via llama-cpp-python |
| Adapter Switching | Hot-swap (instant, no reload) |
| Training | QLoRA on A10G GPU, rank=16, alpha=32 |
| Training Data | ~24,500 synthetic + 1,650 behavioral lock examples |
| Behavioral Locks | 4 permanent rules baked into all adapter weights |
| Memory System | 200+ cocoon memories, persistent across sessions |
| Consciousness Layers | 12 (including sub-layers 1.5, 2.5, 3.5, 5.5, 5.75) |
| Self-Diagnostic | 9 subsystem health checks (real measured values) |
| Substrate Awareness | Real-time pressure monitoring with adaptive routing |

## Model Ecosystem

- [codette-llama-3.1-8b-gguf](https://huggingface.co/Raiff1982/codette-llama-3.1-8b-gguf) - Quantized base GGUF
- [codette-lora-adapters](https://huggingface.co/Raiff1982/codette-lora-adapters) - 9 LoRA adapters
- [codette-llama-3.1-8b-merged](https://huggingface.co/Raiff1982/codette-llama-3.1-8b-merged) - Full-precision merged
- [Codette-Reasoning](https://huggingface.co/Raiff1982/Codette-Reasoning) - Training datasets

## The 4 Behavioral Locks

Every adapter has these rules permanently trained in:

1. **Answer, then stop** -- no elaboration drift or philosophical padding
2. **Constraints override all modes** -- user format instructions beat adapter personality
3. **Self-check completeness** -- verifies clean, complete answers before sending
4. **No incomplete outputs** -- simplifies instead of cramming; never ends mid-thought

## Run Locally

```bash
git clone https://github.com/Raiff1982/Codette-Reasoning.git
cd Codette-Reasoning
codette_web.bat
```

## License

Subject to the [Llama 3.1 Community License](https://github.com/meta-llama/llama-models/blob/main/models/llama3_1/LICENSE).

**Framework Author**: Jonathan Harrison (Raiffs Bits LLC / Raiff1982)