// the find
melgarafael/DeskcommCRM
Open-source AI sales OS — self-hosted CRM with native AI agents + WhatsApp (WAHA). Open alternative to Kommo, Octadesk & Intercom for any business that sells by chat. MCP-ready, multi-tenant, LGPD.
DeskcommCRM is a self-hosted CRM built around WhatsApp selling — kanban pipeline, AI agents with per-tenant RAG that qualify leads and hand off to humans, and either WAHA (QR-based) or the official Meta Cloud API as the WhatsApp channel. It's aimed at Brazilian SMBs (clinics, real estate, e-commerce) who sell over chat and want an alternative to Kommo/Octadesk without a subscription.
The RLS multi-tenant isolation test is done properly: it creates two orgs, drives the same auth.uid()/fn_user_org_ids() path production policies use, and includes a control case proving org B's rows actually exist before asserting org A can't see them. The hostgator-setup-kit install/update flow is more careful than typical self-host scripts — idempotent, takes a DB backup before every update, and auto-rolls-back the container image if a new version fails health checks. CI actually gates on schema idempotency (baseline.sql applied in install mode then update mode, both with ON_ERROR_STOP=1) rather than just running unit tests.
The migration chain is broken by design — migrations 0001-0009 and 0013 are literal `SELECT 1;` stubs, and the real schema only exists in a single baseline.sql. That means no usable migration history and `supabase db push` silently produces an empty database, which will trip up anyone used to normal Supabase/Postgres workflows. The primary README is structured around a paid HostGator VPS affiliate link as the main install path, which reads more like a hosting funnel than an open-source quickstart. WhatsApp automation over WAHA/QR is a ToS gray area with Meta — throttling and jitter reduce but don't remove ban risk, which is a real problem for something positioned as a sales-critical system. The surface area is large for a self-hoster to actually operate (196 route handlers, background workers, multi-provider AI routing, webhook/automation engine), and the docs don't make clear how much ongoing maintenance burden that implies once WAHA sessions or workers misbehave in production.