Most teams that run customer conversations on WhatsApp end up juggling a SaaS inbox, a database, and separate automation tooling — and lose control of data and AI costs. This template trades a managed product for an opinionated, fork-and-run CRM that integrates the official WhatsApp Business API, Supabase-backed data, and an LLM assistant that you power with your own OpenAI or Anthropic key.
What Sets It Apart
- Bring-your-own-key AI assistant: admins paste an OpenAI or Anthropic API key (stored AES-256-GCM encrypted) so the instance calls the provider directly — no per-seat AI fees and no third‑party AI proxying. Draft replies in the inbox or enable a capped auto-reply bot (per-conversation limit and clear human handoff).
- Grounded retrieval + hybrid search: a built-in knowledge-base manager supports lexical Postgres full-text search for every install; adding an embeddings key enables pgvector semantic search (OpenAI embeddings) for higher-quality retrieval while falling back to lexical results when needed. Anthropic-only setups keep lexical search without extra setup.
- Production-ready, fork-first approach: shipped as a template (MIT) to fork, customise and host anywhere Node.js runs. The repo includes security primitives (RLS, HMAC-verified webhooks, token encryption), a visual no-code flow builder, broadcasts with Meta-approved templates, and a public REST API for automations.
Who It's For + Tradeoffs
Great fit if you need full ownership of data and AI credentials, want a turnkey WhatsApp inbox + CRM combining conversations with sales pipelines, and are comfortable self-hosting (or using a recommended Hostinger deploy). Look elsewhere if you need a managed SaaS with hosted AI billing, if you prefer a no-maintenance cloud product, or if you require out-of-the-box enterprise SLAs: this repo expects you to provision Supabase and a WhatsApp Business API setup and to manage hosting and secrets.
Where It Fits
Technically it's an application-layer AI client inside a CRM: the assistant enhances agent workflows (drafting replies, auto-reply with human handoff) rather than being a standalone LLM product. It pairs well with teams that already use Supabase/Postgres and want to add retrieval-augmented LLM drafts and bounded automation to a WhatsApp-centric support or sales workflow.
