AgenticFlow Orchestrator

Enterprise-Grade AI Service Mesh & Multi-Agent Orchestrator

🌐 Live Production URL: https://agenticflow.scholarme.in/

AgenticFlow is a highly scalable, fully containerized AI orchestrator that dynamically routes complex intents to specialized AI agents. Built entirely on a modern service mesh architecture, it securely isolates user data, maintains strict access control, and delivers state-of-the-art responses through Hybrid Retrieval (Vector + Graph), Web Research, and dynamic SQL querying.

---

🚀 Key Innovations & Features

1. 🧠 Autonomous Multi-Agent Routing (LangGraph)

Unlike standard linear chatbots, AgenticFlow uses a sophisticated state machine built with LangGraph. User queries are automatically analyzed and routed to the most capable agent:

• RAG Agent: Ingests and synthesizes information from uploaded documents.
• SQL Agent: Analyzes requirements and dynamically generates valid SQL queries (without direct execution against production databases).
• Research Agent: Conducts live web research via Tavily API to fetch real-time data.
• Code Agent: Executes sandboxed Python code for complex mathematical or programmatic logic.

2. 🗄️ Hybrid Knowledge Engine (pgvector + Memgraph)

AgenticFlow implements a dual-database intelligence layer:

• PostgreSQL (pgvector): Handles semantic similarity search, dense vector embeddings, and persistent chat histories.
• Memgraph (Knowledge Graphs): Maps complex entity relationships, allowing the LLM to traverse highly connected data points that standard vector search misses.

3. 🔐 Enterprise Security & Data Isolation

Security is built into the foundation of the orchestrator, ensuring zero cross-tenant data leakage:

• Strict User-Based Isolation: Every document embedded, every SQL query executed, and every chat session is strictly partitioned by user_id. A user can never access another user's vector space.
• JWT Authentication: Cryptographically secure login and registration utilizing bcrypt hashing and HTTP Bearer tokens.
• Guest Sandboxing: Unauthenticated visitors receive a unique X-Guest-ID. Their sessions are aggressively rate-limited (e.g., 5 messages max) and completely isolated in memory before prompting for account creation.

4. 🎨 Premium Modern Frontend

The UI isn't just an afterthought—it's a massive competitive advantage:

• Built on Next.js (React 18) for blazing-fast SSR and hydration.
• Features a custom Warm Charcoal & Sophisticated Orange aesthetic with smooth glassmorphism, micro-animations, and dynamic gradients.
• Thread-safe Execution: The UI operates independently, communicating with the heavy LangGraph nodes purely via async REST APIs to prevent event-loop blocking.

---

🏗️ Architecture Mesh

The entire system runs as a multi-container Docker mesh, ensuring exact parity between local development and AWS production.

graph TD
    UI[Next.js Premium UI <br/> :3005] -->|REST / JSON| GW(FastAPI Gateway <br/> :8005)
    
    GW -->|Validate JWT / X-Guest-ID| Auth{Security Layer}
    Auth -->|Route| ORCH[LangGraph Orchestrator]
    
    ORCH --> RAG[RAG Agent]
    ORCH --> SQL[SQL Agent]
    ORCH --> WEB[Research Agent]
    
    RAG <-->|Dense Vectors| PG[(PostgreSQL + pgvector <br/> :5432)]
    RAG <-->|Entity Traversal| MG[(Memgraph <br/> :7687)]
    
    style UI fill:#ea580c,stroke:#c2410c,stroke-width:2px,color:#fff
    style GW fill:#009688,stroke:#00796B,stroke-width:2px,color:#fff
    style ORCH fill:#1C3C3C,stroke:#000,stroke-width:2px,color:#fff
    style PG fill:#336791,stroke:#234a66,stroke-width:2px,color:#fff
    style MG fill:#f15a24,stroke:#c0481c,stroke-width:2px,color:#fff

Loading

---

🐳 Production Deployment (AWS EC2)

The application is engineered for horizontal scaling and currently runs on a production AWS EC2 t3.medium instance. Traffic is securely reverse-proxied providing TLS termination and enterprise-ready network mapping, never exposing raw container ports to the public web.

Docker Services Provisioned:

1. agenticflow-backend: FastAPI + LangGraph worker
2. agenticflow-frontend: Next.js Standalone UI
3. agenticflow-postgres: PostgreSQL with pgvector
4. agenticflow-memgraph: High-performance Graph DB

---

🛠️ Local Development (Quickstart)

Prerequisites

• Docker & Docker Compose v2+
• Google Gemini API Key

1. Clone & Setup

git clone https://github.com/Harsh-Sharma29/AgenticFlow.git
cd AgenticFlow

2. Environment Variables

Create a .env file in the project root:

# ── Security ──────────────────────────────────────────────────────────
JWT_SECRET=super-secure-production-key-here

# ── AI Keys ───────────────────────────────────────────────────────────
GOOGLE_API_KEY=your-gemini-api-key
TAVILY_API_KEY=your-tavily-search-key

# ── System Defaults ────────────────────────────────────────────────────
PRIMARY_LLM_MODEL=gemini-2.5-flash
EMBEDDING_MODEL=gemini-embedding-001
DEBUG=false

3. Launch the Mesh

docker-compose up --build -d

The backend includes a dependency health-check; it waits for both Postgres and Memgraph to be fully ready before spinning up the API. Once the API is healthy, the frontend unlocks.

Service	Container	Host Address & Port	Description
Frontend UI	nexus-frontend	http://localhost:3005	Premium Next.js Web Interface
Backend API	nexus-backend	http://localhost:8005/docs	FastAPI Swagger & REST Endpoints
PostgreSQL	nexus-postgres	localhost:5432	Relational DB + pgvector index
Memgraph	nexus-memgraph	localhost:7687	Bolt protocol port for Knowledge Graph
Memgraph Lab	nexus-memgraph	localhost:7444	HTTP WebSocket port for Memgraph UI

---

Built with ❤️ by Harsh Sharma