💬 iDEVconn AI Chat - Backend Module

@idevconn/ai-chat-be is a drop-in, highly-optimized NestJS module that provides an instant, docs-aware AI customer support backend.

No vector database to host, no complex prompt engineering required. Simply point the module at a folder containing your Markdown documentation, and it handles the rest.

🌟 Core Superpowers

⚡ Zero-DB RAG (Retrieval-Augmented Generation)

We completely eliminated the need for external Vector Databases. On application boot, the module reads your local markdown files, generates vector embeddings, and stores them in a lightning-fast, secure in-memory indexing layer.

🛡️ Enterprise-Grade Security Pipeline

Every user input and LLM output goes through a strict, multi-stage security pipeline:

• Prompt Injection Guard: Advanced pattern matching and Unicode deconfusion blocks exploits.
• PII Protection: Automatically scans for and redacts sensitive information (Emails, SSNs, Credit Cards, IPs).
• Cosine Intent Routing: Compares user input embeddings to your document index to identify off-topic questions, automatically redirecting them before they even hit the LLM.

🔌 Multi-Provider Support

Instantly swap between industry-leading LLMs by just changing an environment variable. We natively support:

• Google Gemini (gemini-2.5-flash)
• OpenAI (gpt-4o)
• xAI Grok (grok-3-mini)

🚀 Quick Start Guide

1. Installation

Install the backend module and the required NestJS throttler:

npm install @idevconn/ai-chat-be @nestjs/throttler

2. Register the Module

Import AiChatModule into your root AppModule:

import { Module } from '@nestjs/common';
import { AiChatModule } from '@idevconn/ai-chat-be';

@Module({
  imports: [
    AiChatModule.forRoot({
      // ─── LLM Provider (required) ───────────────────────────
      provider: process.env.AI_CHAT_PROVIDER || 'gemini',
      apiKey: process.env.AI_CHAT_PROVIDER_API_KEY,

      // ─── RAG Settings (optional) ───────────────────────────
      rag: {
        docsPath: './chatdocs', // Folder containing your markdown files
      },

      // ─── Widget Authenticity ───────────────────────────────
      auth: {
        strategy: 'api-key',
        key: process.env.AI_CHAT_WIDGET_API_KEY, // Shared secret with frontend
      },
    }),
  ],
})
export class AppModule {}

3. Add your Markdown Documents

Create a chatdocs/ directory at your project root and drop your markdown guides, help pages, or FAQs in it. The RAG pipeline will index them automatically:

your-nest-app/
├── chatdocs/
│   ├── getting-started.md
│   ├── advanced-features.md
│   └── pricing-faq.md
├── src/
│   └── app.module.ts
└── .env

4. Setup Environment Variables

# Set to 'gemini', 'openai', or 'grok'
AI_CHAT_PROVIDER=gemini
AI_CHAT_PROVIDER_API_KEY=your_api_key_here

# Used to authenticate requests from your frontend widget
AI_CHAT_WIDGET_API_KEY=your_secure_shared_secret

⚙️ Configuration Reference

Provider Options

| Configuration Option | Environmental Variable | Purpose | Default | | :------------------- | :--------------------------------- | :------------------------------ | :----------------------- | | provider | AI_CHAT_PROVIDER | Swaps LLM strategy | 'gemini' | | apiKey | AI_CHAT_PROVIDER_API_KEY | API Key for the chosen provider | (Required) | | model | AI_CHAT_PROVIDER_MODEL | Specific LLM model selection | 'gemini-2.5-flash' | | embeddingModel | AI_CHAT_PROVIDER_EMBEDDING_MODEL | Embedding model override | 'gemini-embedding-001' |

RAG Options

| Configuration Option | Environmental Variable | Purpose | Default | | :------------------- | :---------------------- | :--------------------------- | :--------- | | rag.docsPath | AI_CHAT_DOCS_PATH | Path to your markdown folder | './docs' | | rag.chunkSize | AI_CHAT_CHUNK_SIZE | Words per vector chunk | 250 | | rag.overlapSize | AI_CHAT_CHUNK_OVERLAP | Overlap to maintain context | 50 |

Security & Limits

| Configuration Option | Environmental Variable | Purpose | Default | | :---------------------------- | :--------------------- | :------------------------- | :------ | | limits.maxRequestsPerMinute | AI_CHAT_MAX_RPM | Rate limit per IP address | 10 | | limits.maxInputTokens | AI_CHAT_MAX_INPUT | Hard limit on prompt size | 2000 | | limits.maxOutputTokens | AI_CHAT_MAX_OUTPUT | Maximum length of AI reply | 1000 |

🔗 Pair with the Frontend Widget

This backend is designed to pair perfectly with our universal, framework-agnostic frontend widget. Check out @idevconn/ai-chat-client to complete the integration!