DeskcommCRM: architecture lessons and AI sales workflows

This Portuguese-language walkthrough explains how DeskcommCRM's creator moved from an overloaded CRM project to a sales system planned around its purpose. He describes the earlier TomikOS project, then discusses feature accumulation, conflicting coding sessions and architectural decisions that became expensive to change. The central lesson is to define the system's purpose, elements and connections before expanding its features.

DeskcommCRM connects a WhatsApp channel, AI agents, customer records and sales pipelines. The demo describes agent tools for recording customer details, checking purchase history and opening cases that humans can resolve. An unusually large order illustrates why an agent needs an escalation path when its instructions do not contain an answer. Configurable follow-up workflows resume unanswered conversations and stop when the customer replies. Human handoff, generated website forms and intent routing connect the same sales process.

The setup section covers SSH access to a VPS and the repository's installer command. The installer requests a domain, an SSL email address, Supabase credentials and an AI provider API key. OpenRouter, Anthropic and OpenAI are offered as provider choices. The application runs on the user's server, while the demonstrated AI calls use external services. The presenter also explains how to choose a name for a white-label deployment.