AI

WhatsApp Knowledge Base + RAG

91% accuracy. 0.8% hallucination. Auto-prompt generator drafts system prompt from your content.

  • No credit card
  • Meta Cloud API
  • India-hosted
  • DPDP compliant
WhatsApp Knowledge Base + RAG

Ground every WhatsApp AI reply in your own content — RichAutomate Knowledge Base uses pgvector retrieval-augmented generation (RAG) over your PDFs, FAQs, URLs, and policy docs. Auto-prompt generator drafts the system prompt from your content sample. BYOK embeddings + chat (OpenAI / Google / Sarvam). Per-tenant isolation, encrypted at rest, DPDP-compliant.

What the Knowledge Base does

Upload PDFs, paste URLs, or sync text. Documents are chunked (500 tokens + 80-token overlap), embedded (1536-dim vectors via your chosen provider), stored in Postgres pgvector. On every user message, top-K=5 chunks retrieved, injected into the LLM prompt alongside system instructions + last 6 conversation turns. Answer cites context; refuses to invent.

Real Indian cohort numbers

MetricPre-RAG botRichAutomate KB
Answer accuracy62%91%
Hallucination rate14%0.8%
Repeat-question rate34%9%
Cost / 1k conversations₹520+₹350 (BYOK)

What you can put in it

  • Product catalogs + SKU descriptions + price sheets
  • Policy docs (refund, shipping, terms, T&C)
  • Onboarding manuals + FAQ libraries
  • Internal SOPs for agent-assist + escalation
  • Compliance docs (DPDP, GDPR, sector regulations)
  • Live URL crawls (auto-refresh weekly)

How it works (3 steps)

  1. Create a Knowledge Base at Knowledge Base, bind to your AI Provider credential.
  2. Upload docs (PDFs, TXT, URLs) — auto-chunked + embedded.
  3. Test in playground, then drag the AI Chatbot node into a flow, bind KB.

vs alternatives

CapabilityRichAutomateOpenAI AssistantsCustom RAG stack
WhatsApp-nativeYesNo (need glue)Build it yourself
BYOK5 providersOpenAI onlyAny
Vector DBpgvector (managed)OpenAI-managedPinecone/Weaviate/Chroma
Auto-prompt generatorYesNoBuild it
Time-to-ship1 hour1-2 days2-4 weeks

Pricing

Knowledge Base + AI Chatbot bundle: ₹2,000/month per tenant. Embedding + chat costs billed direct to your provider (BYOK) — typically ₹0.04 embed + ₹0.30 chat per query.

FAQs

FAQ

Questions we get a lot.

Still curious? Reach the team on WhatsApp — we reply within 2 hours during India business hours.

What document types does the Knowledge Base support?

+

PDF, TXT, MD, CSV up to 25 MB per file. URL crawls (auto-refresh weekly). DOCX + XLSX support ships Q3 2026.

How does the Knowledge Base chunk documents?

+

500-token chunks with 80-token overlap, breaking on semantic boundaries (\n\n preferred, then ". ", then char split). Deduped by SHA256 hash. You can review chunks before publishing.

Which embedding models are supported?

+

OpenAI text-embedding-3-small (default, 1536-dim), Google Gemini embedding-001, Sarvam embeddings for Indic. Anthropic does not embed — pair Anthropic chat with OpenAI/Google embeddings.

Is the Knowledge Base data shared across tenants?

+

No — per-tenant isolation enforced at the database level. Vectors stored under tenant_id; queries scoped via Laravel's middleware + SQL where-clauses. Cross-tenant access is impossible by design.

How is the system prompt auto-generated?

+

Paste a 50-20,000 char sample of your content + your KB name. RichAutomate sends it to your bound LLM with a meta-prompt asking for a tailored 150-300 word system prompt that identifies persona, key topics, honesty rules, response style, and grounding constraints. You can edit before saving.

Ready to ship
knowledge base + rag?

14-day free trial. Cancel anytime. Meta Cloud API. India-hosted. DPDP-compliant.

WhatsApp Knowledge Base RAG · pgvector · BYOK Embeddings India 2026 | RichAutomate