Debo — Your Life Intelligence System

Debo is not a journal with a chat box. It stores meaning, not just text — turning your life data into genuine self-knowledge.

What is Debo?

Debo is a Multimodal Intelligence Lab that turns your writing, research, and data into a private memory graph. We blend Personal Research with Collaborative Intelligence, allowing you to not just store data, but to train your own personal AI on your life.

Core Features

• Multimodal Foundation — Processes text, voice, images, and research papers (PDFs) into a unified knowledge base.
• Tinker API — Researcher-grade control over your personal model training and fine-tuning (LoRA).
• Collaborative Intelligence — A partner that thinks with you, citing your own history and data for every insight.
• Connectionism Engine — Discover latent manifolds and hidden relationships across your memory graph.
• Character Graph — Turn people mentioned in chat, journals, and voice into editable profiles with relationships and source references.
• Voice Capture (Jarvis) — Real-time ambient voice interface with sub-second response powered by NVIDIA.

Tech Stack

• Framework: Next.js 16 (App Router), React 19
• Orchestration: Mastra (multi-agent system)
• AI: NVIDIA NIM / OpenAI / Anthropic
• Database: Neon (PostgreSQL), Qdrant (Vector DB)
• Media Storage: Google Drive (video/audio journals)
• Auth: Stack Auth
• Voice: LiveKit
• Deployment: Cloudflare Workers

📋 See Project Roadmap for feature status and upcoming work

Getting Started

# Install dependencies
bun install

# Copy environment variables
cp .env.example .env.local

# Set up database
bun run db:push

# Run development server
bun run dev

Then open http://localhost:3000

Project Structure

src/
├── app/           # Next.js App Router pages
├── actions/       # Server Actions (auth, chat, journals, memories)
├── components/    # UI components (dashboard, landing, chat)
├── db/            # Drizzle ORM schemas
├── features/      # Scalable product modules such as Characters
├── lib/           # Core logic (AI, vector search, memory)
└── mastra/        # Mastra AI agents

Philosophy

Debo follows Editorial Calm — a design philosophy focused on:

• Distraction-free, magazine-like aesthetic
• Typography-first approach
• Minimal UI with generous whitespace

Technical Architecture

graph TD
    User[User] --> Web[Next.js 16 Web UI]
    User --> Voice[LiveKit Jarvis Voice Agent]
    
    subgraph Edge[Cloudflare Edge]
        Web --> Actions[Server Actions]
        Web --> ChatRoute[Chat Runtime]
    end

    subgraph Orchestration[Mastra]
        ChatRoute --> Agents[AI Agents]
        Agents --> Tools[Tools]
    end

    subgraph Storage[Data Layer]
        Tools --> DB[(Neon Postgres)]
        Tools --> Qdrant[(Qdrant Vector)]
    end

    Agents --> LLM[LLM Provider]
    LLM --> Stream[Streaming Response]
    Stream --> User

Loading

Data Flow

1. Journal Entry → Saved to Neon Postgres
2. Embedding → Chunked and stored in Qdrant for semantic search
3. Memory Extraction → Facts, preferences, people stored in memory table
4. Character Sync → People are deduplicated into editable character profiles with references
5. Query → Retrieved from both Qdrant (semantic) + Postgres (structured)
6. Answer → Grounded response with citations sent back to user

License

MIT © SH20RAJ