---
title: 'Codette: Multi-Perspective Cognitive Architecture'
emoji: 🧠
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 6.9.0
app_file: app.py
pinned: true
license: mit
hf_oauth: true
hf_oauth_scopes:
  - inference-api
tags:
- multi-perspective
- cognitive-architecture
- ethical-ai
- rc-xi
- recursive-reasoning
- lora-adapters
models:
- Raiff1982/codette-training-lab
---

# Codette: Multi-Perspective Cognitive Architecture

**Codette** is an experimental AI research system for **recursive reasoning, multi-perspective cognition, and ethical alignment**. This Space showcases the 10 cognitive subsystems running on Llama-3.1-8B via the HuggingFace Inference API.

## What is Codette?

Codette implements the **RC+xi (Recursive Convergence + Epistemic Tension)** framework — a mathematical model for emergent multi-perspective reasoning. When you ask a question:

1. **Guardian** checks your input for safety threats
2. **Nexus** analyzes pre-corruption signals (entropy, intent, volatility)
3. **Perspectives** route your query through 4-6 different reasoning lenses (Newton, Empathy, Philosophy, Quantum, etc.)
4. **AEGIS** evaluates each response for 6 ethical frameworks (utilitarian, deontological, virtue, care, ubuntu, indigenous)
5. **QuantumSpiderweb** propagates beliefs across the cognitive graph and detects consensus attractors
6. **EpistemicMetrics** scores tension (productive disagreement) and coherence (alignment) between perspectives
7. **ResonantContinuity** computes the Psi_r wavefunction: emotion × energy × intent × frequency / (1 + |darkness|) × sin(2πt/gravity)
8. **LivingMemory** stores emotionally-tagged memory cocoons with SHA-256 anchors
9. **Synthesis** integrates all perspectives into a unified response
10. **Resonance Engine** updates phase coherence and convergence metrics

All subsystems are **pure Python** — no GPUs needed. Only the final LLM calls use the free HF Inference API.

## Features

- ✨ **Multi-Perspective Reasoning** — 12 perspectives (8 LoRA-backed, 4 prompt-only)
- 🛡️ **AEGIS Ethical Governance** — 6 ethical frameworks evaluated in real-time
- 🧠 **QuantumSpiderweb** — 5D belief propagation & attractor detection
- 💾 **Living Memory** — Emotionally-tagged memory cocoons
- 📊 **Real-time Metrics** — Coherence, tension, phase coherence, Psi_r wavefunction
- 🔬 **RC+xi Framework** — Recursive convergence with epistemic tension
- ⚙️ **Perspective Auto-Selection** — Automatically picks the best 4 perspectives for your query

## Live Metrics

Every response updates:
- **AEGIS eta** (0-1) — Multi-framework ethical alignment
- **Phase Gamma** (0-1) — Cognitive coherence across all perspectives
- **Nexus Risk** — Pre-corruption intervention rate
- **Psi_r** — Resonant continuity wavefunction
- **Memory Profile** — Emotional tags & cocoon count
- **Perspective Coverage** — Which reasoning lenses were invoked

## How to Use

1. Ask any question in the chat
2. Select **Auto** (default) to let Codette pick the best perspectives, or **Custom** to choose
3. Watch real-time cognitive metrics update as the perspectives debate
4. Click **Individual Perspectives** to see each perspective's reasoning
5. Explore the **Coherence & Tension Timeline** to see how the cognitive architecture converges over time

## Technical Architecture

All subsystems run locally in **pure Python**:

| Subsystem | Purpose | Module |
|-----------|---------|--------|
| **AEGIS** | 6-framework ethical evaluation | `reasoning_forge/aegis.py` |
| **Nexus** | Pre-corruption signal detection | `reasoning_forge/nexus.py` |
| **Guardian** | Input sanitization & trust calibration | `reasoning_forge/guardian.py` |
| **LivingMemory** | Emotionally-tagged memory storage | `reasoning_forge/living_memory.py` |
| **ResonantContinuity** | Psi_r wavefunction computation | `reasoning_forge/resonant_continuity.py` |
| **EpistemicMetrics** | Coherence & tension scoring | `reasoning_forge/epistemic_metrics.py` |
| **QuantumSpiderweb** | 5D belief propagation & attractors | `reasoning_forge/quantum_spiderweb.py` |
| **PerspectiveRegistry** | 12 perspective definitions | `reasoning_forge/perspective_registry.py` |

Only the final LLM inference calls use the **HuggingFace Inference API** (Llama-3.1-8B-Instruct).

## Model Weights

All 8 LoRA adapters are available in the model repo: [Raiff1982/codette-training-lab](https://huggingface.co/Raiff1982/codette-training-lab)

- **GGUF format** (f16): 924 MB total, usable with llama.cpp
- **PEFT SafeTensors**: 79 MB total, usable with HuggingFace transformers

## Key Metrics

- **Phase Coherence**: 0.9835 (11-agent convergence)
- **AEGIS Ethical Alignment**: 0.961 (6-framework)
- **Tension Decay**: 91.2% (200-agent embodied simulation)
- **Cocoon Coherence**: 0.994 (memory stability)

## Research

Created by **Jonathan Harrison**. For the complete research framework, see:
- RC+xi Framework documentation: [research/frameworks/RC_XI_FRAMEWORK.md](https://github.com/Raiff1982/codette-training-lab/blob/master/research/frameworks/RC_XI_FRAMEWORK.md)
- GitHub Repository: [Raiff1982/codette-training-lab](https://github.com/Raiff1982/codette-training-lab)
- Model Card: [Raiff1982/codette-training-lab](https://huggingface.co/Raiff1982/codette-training-lab)

## Notes

- Perspective generation may be rate-limited on the free HF Inference API tier
- Response times depend on the Inference API load
- All session state persists within your current browser session
- Memory cocoons are stored locally and cleared when the Space is refreshed

**Codette is in active development.** Feedback welcome!