OpenClaw: WhatsApp-Native Agentic LLM Orchestration for Multi-Domain Small-Business Operations — A Three-Instance Production Case Study
Vibhav Aggarwal · Zenodo (CERN European Organization for Nuclear Research) · 2026
{WhatsApp} is the de facto operating system for Indian small and medium businesses: orders flow through DMs, payments are confirmed in groups, supplier coordination happens in broadcast lists, and employees already have the application open all day. Yet the agentic-LLM literature has overwhelmingly standardized on two interface modalities — the web chat tab and the in-terminal coding assistant — neither of which meets the Indian SMB operator where they already are. We describe OpenClaw, an agentic-LLM orchestration runtime deployed in production across three instances for two small businesses in Haryana, India: an Ayurvedic D2C retail brand (Herbilé, operated by the author) and a gasket manufacturing company (Genauto Gasket LLP, operated by the author's consultancy client). The three instances span a Raspberry Pi 5 (ARM64, office), an Intel i7-6700 desktop (x86, office), and an Intel i7-12700F workstation (x86, factory). Each instance is a WhatsApp-native gateway that fronts a multi-model LLM stack (Claude Opus 4.6 primary, with a 32-model fallback chain), a persistent-memory layer based on Mem0 + Qdrant + local Ollama embeddings, a systemd-managed service footprint including daily backups, and a ticketing dashboard (“Mission Control”) that turns ad-hoc WhatsApp requests into tracked work items. We describe the architecture, the deployment decisions that survived contact with production, and the engineering gotchas — including ARM64 jemalloc page-size crashes in Qdrant on the Raspberry Pi, a hardcoded mem0_store hostname that required patching the Mem0 Python module, and the dual-SDK headache of keeping npm-global symlinks consistent across three different Linux distributions. We position the work as an existence proof that capable agentic-LLM orchestration does not require SaaS platform lock-in or dedicated UI surfaces, and that the cheapest-off-the-shelf chat application is a productive deployment target.