A textile wholesaler in Ernakulam was spending three hours daily answering the same WhatsApp queries: minimum order quantities, available fabric colours, delivery timelines to different districts, GST invoice requirements. His catalogue had 340 products. Training a conventional chatbot on all of that would have cost ₹80,000 and taken months. Instead, he set up a RAG-based support system in a weekend for under ₹3,000 in total costs. His team now handles the exception queries; the bot handles everything in the catalogue.
RAG — retrieval-augmented generation — is the architecture that makes this possible. Understanding what it does, and more importantly what the no-code tools built on top of it actually require from you, is the first step to deciding whether it fits your business.
What RAG Actually Does (Simply)
A conventional AI chatbot has knowledge baked into it during training. It can tell you about the world in general but knows nothing about your specific product catalogue, your return policy, or your pricing. To change what it knows, you retrain it — an expensive, slow process.
RAG works differently. When a customer types a question, the system first searches your uploaded documents for the most relevant passages, then sends those passages plus the question to an AI model for answering. The model sees: "Here are the relevant bits from the business's own documents. Now answer this question using that information." The knowledge isn't baked in — it's looked up fresh each time.
This means two things that matter enormously for Indian SMBs. First, updating your bot's knowledge requires only updating a document — upload a new price list and the bot immediately knows the new prices, no retraining involved. Second, the bot can only say things that are in your documents, which dramatically reduces hallucination. It won't invent a return policy that doesn't exist or promise a delivery timeline you can't meet.
No-Code RAG Tools Worth Using in India
Botpress has the most generous free tier of any serious RAG platform — their free plan covers up to 2,000 incoming messages per month and allows knowledge base uploads without any coding. The interface is drag-and-drop, with a visual flow builder for designing conversation paths. For a business handling under 60 support queries per day, the free plan is sufficient to start. Paid plans begin at approximately $50/month (around ₹4,200), which includes increased message volume and WhatsApp Business API integration support.
Flowise is open-source and self-hostable — meaning you can run it on your own server for a one-time setup cost with no ongoing per-message fees. A basic VPS on Hostinger India or DigitalOcean running Flowise costs ₹1,200–1,800 per month. The interface uses a node-based visual builder where you connect components (document loader → text splitter → embedding model → vector store → LLM → output). It's more technical than Botpress but gives full control over which models you use, which is relevant if you want to avoid per-token cloud AI costs.
Dify.ai sits between the two in complexity. Its cloud version has a free tier for small deployments and a more streamlined setup process than Flowise. The knowledge base management interface is particularly clean — you upload PDFs and documents through a simple file manager, set chunking parameters, and the system handles the rest. For businesses that want something operational in hours rather than days, Dify is often the fastest path.
n8n is primarily a workflow automation tool rather than a dedicated RAG platform, but its AI agent nodes can orchestrate RAG pipelines effectively. If you're already using n8n for other automation (WhatsApp message handling, CRM updates, order notifications), adding a RAG step to an existing workflow is straightforward. Self-hosted n8n costs nothing for the software; you pay only for your VPS and any AI API calls.
Preparing Your Documents for RAG
The quality of a RAG system's answers is almost entirely determined by the quality of the source documents. Poorly structured documents — scanned PDFs with image-only content, spreadsheets with merged cells and unclear column headers, informal notes written for internal use — produce unreliable answers.
For Indian businesses setting up their first RAG system, the most useful document types are:
- Product catalogues in PDF or Word format — with clear product names, SKU codes, specifications, prices (including GST-inclusive and exclusive), and availability status
- FAQ documents — written specifically to be read by an AI, meaning questions and answers formatted clearly rather than embedded in flowing prose
- Policy documents — return policy, warranty terms, shipping timelines by region, payment methods accepted (UPI, NEFT, Razorpay, COD)
- Service descriptions — what each service includes, what it doesn't, pricing tiers, turnaround times
A common preparation mistake: uploading documents written for customers alongside documents written for internal staff. Internal documents often contain caveats, codes, and exceptions that should never surface in customer-facing answers. Keep these libraries separate.
Embedding Models: What They Are and Which Are Free
The "retrieval" part of RAG requires converting your documents into numerical representations (embeddings) that the system uses for search. Every RAG tool handles this internally, but the model used for embedding affects both cost and accuracy.
OpenAI's text-embedding-3-small model costs $0.02 per million tokens — negligible for a business-sized knowledge base. A 50-page product catalogue embedded once costs under ₹5. You only pay for embedding when documents are added or updated, not on every query.
Free alternatives include Sentence Transformers models that run locally (no API cost at all) and Google's embedding models via Vertex AI (which has a free tier). For Flowise self-hosted deployments, using a local embedding model eliminates the embedding cost entirely — relevant if you're updating your catalogue frequently.
Real Use Case: Kerala Textile Wholesaler
The Ernakulam wholesaler mentioned at the start set up his system using Dify.ai's cloud version. His knowledge base consisted of three documents: a 22-page product catalogue (fabric types, GST HSN codes, price per metre at different MOQs), a shipping policy document covering delivery timelines to all 14 Kerala districts and major Gulf shipping agents, and a two-page FAQ covering common queries about colour fastness, sample availability, and invoice requirements.
Setup took approximately six hours across two days — mostly spent cleaning up the catalogue document so product information was consistently formatted. The bot was connected to WhatsApp Business API through a BSP, so customer queries that arrived on WhatsApp were handled automatically during non-business hours.
In the first month, the bot handled 68% of incoming queries without escalation. The remaining 32% — mostly questions about custom dyeing, bulk discounts on specific orders, and delivery to non-standard addresses — were automatically routed to the owner's personal WhatsApp number. His team's daily support time dropped from 3 hours to under 45 minutes.
Cost Breakdown: What You Actually Pay
For a small Kerala business handling 30–80 customer support queries per day:
- Botpress free tier: ₹0/month for setup and up to 2,000 messages. WhatsApp integration requires a BSP (Interakt basic plan: ₹2,499/month). Total: ₹2,499/month.
- Dify.ai cloud: Free tier covers modest usage. Paid plans from approximately ₹2,000/month. Add BSP costs for WhatsApp: ₹5,000–8,000/month all-in.
- Flowise self-hosted: VPS ₹1,500/month + OpenAI API costs (typically ₹500–1,500/month at this query volume) + BSP for WhatsApp ₹2,499/month. Total: ₹4,500–5,500/month with full control over costs.
- Enterprise chatbot (traditional vendor): Setup ₹50,000–1,50,000 + ₹8,000–15,000/month maintenance. Retraining for catalogue updates: ₹5,000–20,000 per update.
The economic case is straightforward. The RAG approach costs 30–60% less than a traditional chatbot vendor while offering faster knowledge updates and significantly lower hallucination rates.
Connecting RAG to WhatsApp Business API
Most Indian businesses want their AI support accessible on WhatsApp, which is where their customers already are. This requires a WhatsApp Business API account through a Business Solution Provider (BSP).
BSPs with strong India support and Malayalam-language customer service options include Interakt (popular with Kerala businesses for its Hindi and English UI), Wati (good API documentation for custom integrations), and Zoko (competitive on per-message pricing for high-volume businesses). All three offer webhook-based integration with Botpress, Flowise, and Dify.
The integration flow: customer sends WhatsApp message → BSP receives it → forwards to your RAG system via webhook → RAG system searches knowledge base and generates response → sends reply back through BSP to customer's WhatsApp. The whole round trip takes 2–5 seconds on a typical connection.
When RAG Doesn't Work Well
RAG handles well-defined questions about documented information exceptionally well. It struggles in three specific scenarios worth knowing before you commit.
Very large, constantly changing catalogues: A business with 5,000+ active SKUs that update daily will find RAG retrieval accuracy degrades — the system retrieves the most semantically similar chunks but may miss relevant items buried in large documents. At this scale, you need a proper product database with a structured query layer, not pure RAG.
Real-time data requirements: RAG can't answer "Is order #1234 dispatched yet?" unless your shipping status data is continuously written into the knowledge base or the system has direct database access. Static document RAG is unsuitable for live order tracking — you need a different integration approach for that.
Complex multi-step calculations: Questions like "What would be the total cost for 200 metres of fabric type A, including GST and delivery to Wayanad?" require arithmetic that RAG will sometimes get wrong. For pricing calculators, a rule-based system or a simple web form is more reliable than asking an LLM to do the maths.
For the vast majority of Kerala SMBs — businesses whose customers ask predictable questions about products, services, policies, and availability — RAG-based support using no-code tools is a practical, affordable upgrade that genuinely works. The ceiling on complexity is lower than a custom-coded solution, but for many businesses, the ceiling is comfortably above where their actual support questions live.