Inqora

Inqora is an enterprise-grade cognitive memory framework and multi-agent intelligence monorepo. It seamlessly unifies short-term working context, long-term semantic vector memory, knowledge graphs, VLM visual understanding, cross-platform mobile experiences, and hands-free desktop stealth overlays.

---

🏗️ Architecture & Directory Structure

Inqora is organized as a high-performance monorepo managed via Turborepo and npm workspaces.

23-inqora/
├── apps/
│   ├── web/               # Next.js 16 frontend (Blinky showcase, Chat UI, Document OCR, Panel)
│   ├── blinky/            # Electron + React 19 stealth desktop assistant overlay
│   ├── server/            # Node.js Express backend (Vector memory, SQLite, BullMQ, LangChain)
│   ├── agentos/           # Go-based high-performance agent runtime & CLI execution engine
│   ├── rag-pipeline/      # Python FastAPI microservice for RAG document indexing & search
│   ├── mobile-app/        # Cross-platform Expo / React Native app with NativeWind
│   ├── docs/              # Mintlify documentation portal
│   └── ai-video-editor/   # Next-gen visual media processing suite
├── packages/
│   ├── agent-orchestrator/# Shared multi-agent orchestration primitives
│   ├── ui/                # Shared design system & React UI component library (@repo/ui)
│   ├── eslint-config/     # Workspace-wide ESLint configurations
│   ├── tailwind-config/   # Shared Tailwind CSS design system tokens
│   └── typescript-config/ # Strict TypeScript base configurations
├── agent-factory/         # Pre-configured agent sub-systems (e.g. slack-agent)
└── infra/                 # Docker, task runner, and deployment scripts

---

🌟 Key Application Suites & Capabilities

🌐 1. Web Platform (apps/web)

• Tech Stack: Next.js 16 (App Router), React 19, Tailwind CSS v4, Framer Motion, Shiki, Recharts.
• Blinky Landing Page (/): Ultra-minimalist showcase with smooth micro-animations and high-contrast UI tokens.
• Memory-Aware Chat (/chat): Real-time interface revealing context provenance across working memory, vector stores, and knowledge graphs.
• Document & VLM OCR (/document-ocr): Multimodal document parsing powered by Gemini 1.5/2.5 Flash and LandingAI ADE.
• System Panel (/panel): System status monitoring and configuration management.

🎙️ 2. Blinky Stealth Desktop Overlay (apps/blinky)

• Tech Stack: Electron 34, React 19, Vite, Tailwind CSS.
• Stealth Dashboard: Floating overlay with configurable transparency, click-through mode, and screen-share masking (Zoom/Teams friendly).
• Context Awareness: Active screen screenshot capture giving immediate visual workspace context to AI agents.
• Hands-Free Voice: Real-time Speech-to-Text (STT) and spoken Text-to-Speech (TTS) responses.
• Native System Execution: Launches native local Windows/Mac applications based on conversational context.

⚡ 3. Backend & Agent Microservices

• Core Server (apps/server): Node.js Express service backing multi-tiered vector storage (Pinecone, Qdrant, ChromaDB), SQLite cache, BullMQ job queues, and document processors (pdfkit, mammoth, tesseract.js).
• AgentOS (apps/agentos): Ultra-fast Go engine providing CLI management and low-overhead agent orchestration.
• RAG Microservice (apps/rag-pipeline): FastAPI & Python microservice dedicated to scalable vector chunking and document retrieval.
• Mobile App (apps/mobile-app): iOS & Android client built with Expo Router and NativeWind.

---

🧠 Hierarchical Memory System

1. Short-Term Working Memory: Maintains active conversation windows with automated summarization upon exceeding context bounds.
2. Long-Term Semantic Vector Memory: Dual Pinecone and Qdrant retrieval utilizing access frequency and logarithmic memory decay models.
3. Knowledge Graph Integration: Extracts structural entity identities and social relationships, linking evidence with confidence scores.
4. Visual Memory (VLM): Analyzes photos and screen captures for deep scene descriptions, identity clustering, EXIF geospatial tracking, and neural journal generation.

---

🚀 Getting Started

Prerequisites

• Node.js: >=18.0.0
• npm: >=10.0.0
• Docker & Docker Compose (Optional, for containerized local services)
• Go: >=1.21 (For apps/agentos)
• Python: >=3.10 (For apps/rag-pipeline)

Installation & Environment Setup

1. Clone the repository:

   git clone https://github.com/anishs1207/ai-memory.git
   cd 23-inqora

2. Install Monorepo Dependencies:

   npm install

3. Configure Environment Variables: Create a .env file in the root directory:

   GEMINI_API_KEY=your_gemini_api_key
   DATABASE_URL=postgresql://user:password@127.0.0.1:5432/agentic_db?schema=public
   REDIS_HOST=localhost
   REDIS_PORT=6379

Local Infrastructure with Docker

Launch PostgreSQL, Redis, backend services, and web apps with Docker Compose:

docker-compose up -d

Development Commands

Run monorepo tasks across all applications concurrently via Turborepo:

# Start all dev servers (Web on :3000, Server on :3001, Desktop, etc.)
npm run dev

# Compile production builds across all workspaces
npm run build

# Run ESLint validation
npm run lint

# Execute TypeScript typechecking across all packages
npm run check-types

# Execute unit and integration test suites
npm run test

---

🛠️ Monorepo Package Reference

Workspace Package	Type	Description
apps/web	Web Application	Next.js 16 chat interface, VLM document OCR, and Blinky landing page
apps/blinky	Desktop App	Electron floating overlay assistant with speech & local control
apps/server	Express Backend	Core API, vector retrieval, SQLite storage, and BullMQ queues
apps/agentos	Go Service	High-performance agent execution engine
apps/rag-pipeline	Python Service	FastAPI containerized RAG pipeline
apps/mobile-app	Mobile Client	Expo React Native application
apps/docs	Documentation	Mintlify documentation suite
@repo/ui	Shared Package	Shared React component library
@repo/agent-orchestrator	Shared Package	Agent orchestration primitives

---

📜 License

Distributed under the MIT License. See LICENSE for more information.