The conversational commerce
engine.
Nous is the channel-agnostic engine layer that powers Iris and the channel-specific bots we'll ship after it. One reasoning core, many surfaces — Messenger today, WhatsApp next, more as merchants ask.
Why Nous exists
An engine, not a
chatbot builder.
Most chatbot tools ask you to draw flows. Nous starts from the opposite end — a reasoning core with channel adapters and merchant primitives attached. That's the only way conversational commerce actually works in Bangla, Banglish, and voice notes.
- 🧠
LLM-first, not rule-based
Bangla code-mix, intent routing, conversational state — all LLM-handled. Rule-based bots break the moment a buyer types something off-script. Nous treats every turn as reasoning, not regex.
- 🔌
Channel-agnostic
One engine, many surfaces. Messenger today, WhatsApp next, then Telegram, Instagram, and web shop widgets. Merchants don't rebuild flows when a new channel matters.
- 🛠️
Production primitives built in
pgvector retrieval, courier dispatch hooks (Pathao / Steadfast / RedX), payment hooks (bKash / Nagad), human handoff with full thread context. Not a chatbot toy — the plumbing is in the box.
What runs on Nous
One engine,
many channels.
- Live · private betaIris by CaffinixMessenger AI for f-commerce merchants
- NextWhatsApp channel botSame engine, WhatsApp Cloud API surface
- On the roadmapTelegram, Instagram DM, web widgetAdditional channels as Iris scales
Under the hood
Boring stack.
Sharp choices.
We picked tools that survive — not the latest framework of the week. Each piece earns its place.
- Spring Boot
API + orchestration layer. Boring on purpose — JVM stability, clean tool/channel adapters, fast iteration.
- PostgreSQL + pgvector
Merchant-private retrieval. Each tenant gets isolated embeddings — products, FAQs, past customer history. No cross-tenant leakage.
- Gemini · Claude
Routed by turn complexity. Cheap models for routine turns, larger models when judgment is needed. Keeps margin viable at f-commerce price points.
Capabilities
What the engine
actually does.
Channel adapters, one core
Same conversation engine across Messenger, WhatsApp, Instagram, web. Add a channel without rebuilding the brain.
LLM-first orchestration
No hardcoded flows. Intent + context + tools, every turn. Cheap models for routine turns, big models when judgment is needed.
Merchant-private vector memory
Each merchant gets isolated retrieval — products, FAQs, past customer history. No cross-tenant leakage.
Built for low-cost economics
Designed around the $0.003/conversation reality of f-commerce. Quantized routing keeps margin even at scale.
Building a conversational
commerce channel?
Iris is live. The next channels are queued. If you're a platform or marketplace thinking about conversational ordering, talk to us.