🐬 Dolphin: AI with Persistent Graph Memory

Dolphin is a Neural Memory Graph AI that retains knowledge across conversations.
Unlike standard chatbots that forget everything when you start a new thread, Dolphin builds a Global Knowledge Graph of facts, preferences, and relationships, allowing for hyper-personalized interactions.

🎥 Demo

dolphinDemo.1.mp4

🏗️ Architecture

🚀 Key Features

🧠 1. Global Neural Graph (Cross-Session Memory)

Dolphin doesn't just "remember" the last 10 messages. It builds a permanent Knowledge Graph in Supabase.

• Fact Extraction: Every time you chat, Dolphin extracts entities and relationships (e.g., (User) -[LIVES_IN]-> (San Francisco)).
• Global Access: Start a new chat session, and Dolphin still knows who you are. The graph is global, not session-locked.

🔗 2. GraphRAG & Hybrid Retrieval

Traditional RAG (Vector Search) is often not enough. Dolphin combines it with Graph Traversal.

• The "Mental Map": If you ask about "deadlines", Dolphin doesn't just look for the word "deadline". It traverses the graph: Deadline -> Project Beta -> Team Lead -> Alex.
• Contextual Awareness: It understands the relationships between facts, providing deeper, more intelligent answers.

🕸️ 3. Interactive 3D Visualization

See your AI's brain grow in real-time.

• Live Force Graph: A beautiful, interactive 3D view of all your memories.
• Explorable: Zoom, pan, and hover over nodes to see how facts are connected.
• Visual Debugging: Instantly see if the AI has formed the right connections.

⚡ 4. Synaptic Pruning & Consolidation

Just like a human brain, Dolphin sleeps.

• Efficiency: Background processes (using local Llama 3) analyze the graph to merge duplicate nodes and remove irrelevant noise.
• Long-Term Stability: Keeps the graph clean and efficient as it scales to thousands of memories.

🤖 5. Local & Cloud Flexibility

• Hybrid AI: Use Google Gemini (Cloud) for complex reasoning and Ollama (Llama 3) (Local) for privacy-focused tasks or embeddings.
• Cost Effective: Offload heavy lifting to your local GPU.

---

🛠️ Tech Stack

• Frontend: Vanilla JS (ES Modules), Three.js, 3d-force-graph
• Backend: FastAPI (Python), Uvicorn
• Database: Supabase (PostgreSQL + pgvector)
• AI Orchestration: LangChain
• LLMs: Gemini 1.5 Flash, GPT-4, Llama 3 (via Groq), Ollama (Local)

---

🏁 Getting Started

Follow these steps to set up Dolphin locally.

1. Prerequisites

• Python 3.10+ installed.
• Node.js (optional, for package management if needed).
• A Supabase account.

2. Ollama Setup (Local AI)

Dolphin uses local models for embeddings and high-performance tasks.

1. Download Ollama: Go to ollama.com and download the installer for your OS.
2. Install & Run: Run the installer. Ensure the Ollama app is running in the background (you should see an icon in your system tray).
3. Pull the Model: Open your terminal/command prompt and run:

   ollama pull llama3.2

   This downloads the Llama 3.2 model, which Dolphin uses for local inference.

3. Supabase Setup

1. Create a Project: Log in to Supabase and create a new project.
2. Get Credentials: Go to Project Settings -> API. Copy the Project URL and anon public key.
3. Run Migrations (Database Setup):
   • Go to the SQL Editor in your Supabase dashboard.
   • Open the file supabase/migrations/20260214062656_remote_schema.sql from this repository.
   • Copy the entire content of that file.
   • Paste it into the Supabase SQL Editor and click RUN. This will create all the necessary tables (conversation_logs, graph_nodes, graph_edges, user_memories) and enable the vector extension.

4. Environment Variables

1. Copy .env.example to .env:

   cp .env.example .env

   (Or manually rename the file).
2. Open .env and fill in your keys:

   SUPABASE_URL=your_supabase_url
   SUPABASE_KEY=your_supabase_anon_key
   GOOGLE_API_KEY=your_gemini_api_key
   # Add other keys if using OpenAI/Groq

5. Run the Application

1. Install Dependencies:

   pip install -r requirements.txt

2. Start the Server:

   uvicorn server:app --reload

3. Access the UI: Open http://localhost:8000 in your browser.

---

🧪 Usage

1. Chat: Start talking! Tell Dolphin about yourself.
2. Visualize: Click the "Graph" icon to see your memory grow.
3. Settings: Click the "Settings" icon to switch between Cloud (Gemini) and Local (Ollama) models.

---

🗺️ Roadmap

• Beta 1.0 (Current Version):

  • Full local setup (Supabase + Ollama + Python).
  • Experience the Neural Memory Graph firsthand.
  • Ideal for developers and testers.

• Alpha 1.0 (Upcoming):

  • Dolphin SDK & CLI: pip install dolphin-memory.
  • Easily integrate the memory layer into your existing Python products.
  • Automated setup tools.
    PS: Work on Alpha 1.0 has started !!! Stay tuned for the package release.

• Alpha 2.0 (Future):

  • Dolphin Cloud: A fully managed SaaS platform.
  • We handle the database, graph infrastructure, and scaling.
  • Simple API for developers (like OpenAI/Supabase).

---

Made with ❤️ by DewashishCodes