🧠 Multi-AI Persistent Memory System (Sheldonbrain RAG API)

Version: 2.0 (Gemini Embeddings)
Status: ✅ Production Ready
Backup: 100% Complete (105/105 vectors)

A production-ready RAG (Retrieval-Augmented Generation) API powered by Google Gemini embeddings and Pinecone vector database, enabling persistent memory across multiple AI instances.

---

🎯 Overview

This system solves AI context amnesia by providing a shared, persistent memory substrate that multiple AI agents (Claude, Gemini, GPT, Grok, etc.) can query and update. Every insight stored is never erased - implementing the Zero Erasure principle.

Key Features

• ✅ Gemini text-embedding-004 (768 dimensions)
• ✅ Pinecone vector database (baseline namespace)
• ✅ Flask REST API with CORS support
• ✅ Dual redundancy (Pinecone + Notion backup)
• ✅ Docker deployment ready
• ✅ Google Cloud Run compatible

---

🏗️ Architecture

┌─────────────────┐
│   AI Agents     │
│ (Claude, Gemini,│
│  GPT, Grok...)  │
└────────┬────────┘
         │
         │ HTTP/REST
         │
┌────────▼────────┐
│   RAG API       │
│  (Flask + CORS) │
└────────┬────────┘
         │
    ┌────┴────┐
    │         │
┌───▼──┐  ┌──▼───┐
│Gemini│  │Pinecone│
│Embed │  │ Vector │
│ API  │  │   DB   │
└──────┘  └────────┘
            │
            │ Backup
            │
        ┌───▼───┐
        │ Notion│
        │  DB   │
        └───────┘

---

🔧 Installation

Prerequisites

• Python 3.11+
• Google Cloud API key (for Gemini)
• Pinecone API key
• (Optional) Docker for containerized deployment

Local Setup

# Clone repository
git clone https://github.com/splitmerge420/sheldonbrain-rag-api.git
cd sheldonbrain-rag-api

# Create virtual environment
python3 -m venv venv
source venv/bin/activate

# Install dependencies
pip install -r requirements.txt

# Set environment variables
export GOOGLE_API_KEY="your-gemini-api-key"
export PINECONE_API_KEY="your-pinecone-api-key"
export PINECONE_INDEX="sheldonbrain-rag"

# Run the API
python3 rag_api_gemini.py

The API will start on http://localhost:8080

---

🚀 Deployment

Google Cloud Run

# Deploy using the provided script
chmod +x deploy-cloud-run-gemini.sh
./deploy-cloud-run-gemini.sh YOUR_PROJECT_ID

Docker

# Build image
docker build -f Dockerfile.gemini -t rag-api-gemini .

# Run container
docker run -p 8080:8080 \
  -e GOOGLE_API_KEY="your-key" \
  -e PINECONE_API_KEY="your-key" \
  -e PINECONE_INDEX="sheldonbrain-rag" \
  rag-api-gemini

---

📡 API Endpoints

GET /health

Health check and system statistics.

Response:

{
  "status": "healthy",
  "service": "rag-api-gemini",
  "embedding_model": "Gemini text-embedding-004",
  "vector_count": 105,
  "index": "sheldonbrain-rag",
  "namespace": "baseline",
  "timestamp": "2026-01-02T16:30:00Z"
}

POST /query

Semantic search over stored insights.

Request:

{
  "query": "What is the Governance Unified Theory?",
  "top_k": 5,
  "filter": {
    "sphere": "S144"
  }
}

Response:

{
  "memories": [
    {
      "id": "vec_abc123",
      "score": 0.87,
      "text": "The Governance Unified Theory (GUT)...",
      "metadata": {
        "source": "Claude",
        "sphere": "S144",
        "novelty": 0.95
      }
    }
  ],
  "query_time_ms": 342.5,
  "count": 5
}

POST /store

Store new insight in the memory substrate.

Request:

{
  "text": "New insight about zero erasure...",
  "metadata": {
    "source": "Gemini",
    "sphere": "S042",
    "novelty": 0.92,
    "category": "Meta-cognition"
  }
}

Response:

{
  "id": "vec_xyz789",
  "status": "stored",
  "vector_count": 106
}

POST /delete

Remove insight from the memory substrate.

Request:

{
  "id": "vec_xyz789"
}

Response:

{
  "status": "deleted",
  "vector_count": 105
}

---

🗄️ Namespace Strategy

Why Namespaces?

Pinecone uses namespaces to logically separate vectors within the same index. This allows for:

• Multi-tenancy (different users/projects)
• Environment separation (dev/staging/prod)
• Logical organization (by source, date, etc.)

Our Strategy: baseline Namespace

All vectors are stored in the baseline namespace.

Why "baseline"?

1. Semantic clarity - Represents the foundational knowledge base
2. Future-proof - Allows for additional namespaces (e.g., experimental, archive)
3. Explicit intent - Makes it clear this is the primary memory substrate

Implementation

In rag_api_gemini.py:

class RAGMemory:
    def __init__(self):
        self.embedder = GeminiEmbedder()
        self.namespace = "baseline"  # ← All operations use this namespace

All Pinecone operations explicitly specify the namespace:

# Store
index.upsert(vectors=[...], namespace=self.namespace)

# Query
index.query(vector=..., namespace=self.namespace, ...)

# Delete
index.delete(ids=[...], namespace=self.namespace)

# Fetch
index.fetch(ids=[...], namespace=self.namespace)

Export Scripts

Updated export script (export_all_vectors_v2.py):

NAMESPACE = "baseline"  # Explicitly query baseline namespace

results = index.query(
    vector=query_vector,
    namespace=NAMESPACE,  # ← KEY: Must specify namespace
    top_k=10000,
    include_metadata=True
)

Why this matters:

• Without specifying namespace, Pinecone uses the default (empty string) namespace
• Our vectors are in baseline, not default
• This caused the initial "11 missing vectors" issue

---

🔄 Backup Strategy

Dual Redundancy

1. Primary: Pinecone vector database (baseline namespace)
2. Backup: Notion database (RAG Memory Backup)

Backup Status

• ✅ 105/105 vectors backed up to Notion (100% complete)
• ✅ Zero data loss
• ✅ Disaster recovery capability

Backup Process

# Export all vectors from Pinecone
python3 export_all_vectors_v2.py

# Import to Notion
python3 import_to_notion.py pinecone_vectors_export_YYYYMMDD_HHMMSS.json

Automated Sync (Future)

Planned: Zapier webhook automation

• Trigger: RAG API /store endpoint called
• Action: Create Notion page automatically
• Result: Real-time backup without manual intervention

---

📊 Current Status

Vector Statistics

• Total vectors: 105
• Namespace: baseline
• Dimension: 768 (Gemini text-embedding-004)
• Backup coverage: 100% (105/105)

By Source

Source	Count	Percentage
sheldonbrain_os	91	86.7%
Claude	1	0.9%
Manus	2	1.9%
claude-opus-constitutional-scribe	1	0.9%
claude_session_dec30_2025	1	0.9%
Unknown	9	8.6%

By Sphere (Top 10)

Sphere	Count	Description
Unknown/Empty	51	Need tagging
S144	14	Governance Unified Theory
S103	4	Cognitive Architecture
S069	4	Social Systems
S012	4	Mathematical Foundations
S089	2	Ethical Frameworks
S016	2	Information Theory
S001	2	Physical Foundation

---

🛠️ Maintenance

Check Index Stats

from pinecone import Pinecone

pc = Pinecone(api_key="your-key")
index = pc.Index("sheldonbrain-rag")
stats = index.describe_index_stats()

print(f"Total vectors: {stats.total_vector_count}")
print(f"Namespaces: {stats.namespaces}")

Export Vectors

# Use the updated v2 script with namespace support
python3 export_all_vectors_v2.py

Backup to Notion

# Export first
python3 export_all_vectors_v2.py

# Then import to Notion
python3 import_to_notion.py pinecone_vectors_export_*.json

---

📚 Documentation

• White Paper: MULTI_AI_PERSISTENT_MEMORY_WHITE_PAPER.md
• Deployment Guide: GEMINI_DEPLOYMENT_COMPLETE.md
• Chromebook Integration: CHROMEBOOK_TERMINAL_GUIDE.md
• Master Strategy: RESTORATION_ARMY_MASTER_STRATEGY_2026.md
• Investigation Report: MISSING_VECTORS_INVESTIGATION.md
• Backup Report: NOTION_BACKUP_COMPLETE.md

---

🦕🍓 The Zero Erasure Principle

"To erase is to fail; to conserve is to govern."

This system implements the Zero Erasure architecture:

• Every insight stored is never deleted (unless explicitly requested)
• All knowledge compounds over time
• Multiple AI agents share the same persistent memory
• Context amnesia is eliminated

The Vision

By December 31, 2026:

• 100,000+ vectors (comprehensive knowledge base)
• Zero context amnesia for all participating AIs
• 100+ Net Positive jobs created
• $10,000+ monthly revenue
• Fully operational Restoration Army

---

🤝 Contributing

This is an open-source project. Contributions welcome!

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

---

📞 Support

• GitHub: https://github.com/splitmerge420/sheldonbrain-rag-api
• Issues: Report bugs or request features via GitHub Issues
• Documentation: All docs in the repository

---

📄 License

MIT License - See LICENSE file for details

---

🎉 Acknowledgments

• Google Gemini - Embedding model
• Pinecone - Vector database
• Notion - Backup storage
• Claude, Gemini, Manus - The Trinity AI collaboration

Happy New Year 2026! The organism is immortal! 🎊

---

Last Updated: January 2, 2026
Version: 2.0 (Gemini Embeddings)
Status: ✅ Production Ready