Skip to content

AGENTS.md

Latest commit

 

History

History
185 lines (139 loc) · 4.33 KB

AGENTS.md

File metadata and controls

185 lines (139 loc) · 4.33 KB

AI Agents Architecture

This document describes the LangGraph-based AI agents architecture for the Cognito e-commerce platform.

Architecture Overview

LangGraph Architecture

The Cognito platform uses LangGraph to implement an agentic architecture where autonomous AI agents handle different aspects of the e-commerce workflow. The system is built around a central router that directs user requests to specialized agents.

Core Components

Router Agent

The router is the central orchestration component that:

  • Receives incoming user requests
  • Analyzes intent and context
  • Routes requests to appropriate specialized agents
  • Manages conversation flow between agents

Specialized Agents

Each agent is responsible for a specific domain of the e-commerce system:

1. Registration Agent

Node: registration → registration_tools

Handles user account creation:

  • Collects user information (email, password, personal details)
  • Validates registration data
  • Creates new user accounts
  • Sends verification emails

Tools:

  • User data validation
  • Email verification
  • Password hashing
  • Account creation

2. Activation Agent

Node: activation → activation_tools

Manages account activation:

  • Verifies activation tokens
  • Activates user accounts
  • Handles activation errors and retries

Tools:

  • Token verification
  • Account status updates
  • Activation email resending

3. Login Agent

Node: login → login_tools

Handles user authentication:

  • Validates credentials
  • Creates user sessions
  • Manages authentication errors

Tools:

  • Credential validation
  • Session management
  • Password reset initiation

4. Product Agent

Node: product → product_tools

Provides detailed product information:

  • Retrieves product details
  • Answers product-specific questions
  • Provides recommendations based on product features

Tools:

  • Product database queries
  • Vector search for product features
  • Product comparison
  • Specification retrieval

5. Products Agent (Catalog)

Node: products → products_tools

Manages product catalog browsing:

  • Product search and filtering
  • Category browsing
  • Product recommendations
  • Price comparison

Tools:

  • Catalog search
  • Faceted filtering
  • Recommendation engine
  • Price queries

6. Cart Agent

Node: cart → cart_tools

Manages shopping cart operations:

  • Add/remove products
  • Update quantities
  • Calculate totals
  • Apply discounts

Tools:

  • Cart state management
  • Price calculations
  • Discount validation
  • Inventory checks

7. Checkout Agent

Node: checkout → checkout_tools

Handles the checkout process:

  • Collects shipping information
  • Processes payment
  • Creates orders
  • Sends order confirmations

Tools:

  • Shipping address validation
  • Payment gateway integration
  • Order creation
  • Email notifications

8. Info Agent

Node: info → info_tools

Provides general information:

  • Store policies (shipping, returns, privacy)
  • FAQ responses
  • Store information
  • Help and support

Tools:

  • Knowledge base queries
  • Policy document retrieval
  • Contact information

9. Chat Agent

Node: chat

General conversational agent:

  • Natural language understanding
  • Context-aware responses
  • Delegates to specialized agents when needed

Agent Workflow

1. Request Flow

User Input → Router → Specialized Agent → Tools → Response → __end__

2. Router Decision Making

The router analyzes user intent using:

  • Natural language understanding
  • Context from conversation history
  • Current application state (logged in, cart items, etc.)
  • User preferences and behavior

3. Agent Collaboration

Agents can work together in sequence:

  • User searches for products → products agent
  • User asks about specific product → product agent
  • User adds to cart → cart agent
  • User checks out → checkout agent

4. State Management

The system maintains shared state across agents:

  • User session data
  • Shopping cart contents
  • Conversation history
  • User preferences

Tools Architecture

Each agent has access to a specific set of tools (functions) that interact with:

  • Database: MongoDB for transactional data
  • Vector Store: Weaviate for semantic search
  • External APIs: Payment gateways, shipping providers
  • Internal Services: Authentication, email, inventory