A Brazilian developer did something very Brazilian—he stuffed an entire sales team into a single VPS, made WhatsApp the battlefield, and let AI agents answer customers, qualify leads, and close deals. Then he open-sourced it.
The project is called DeskcommCRM, and it gained 505 stars in a single day. Its positioning is blunt: an open-source AI sales OS, a self-hosted alternative to Kommo, Octadesk, and Intercom.
One Command, One CRM
The pain point of traditional SaaS CRMs isn't lack of features—it's that you pay for someone else's server every month. DeskcommCRM flips the logic: rent a 4GB VPS, run one bash hostgator-setup-kit/install.sh, and the entire CRM (Next.js 16 + Supabase + WhatsApp gateway + AI agent) comes up. HTTPS is auto-configured, the database schema is auto-created, the admin user is auto-generated, and even cron jobs are installed automatically.
This isn't "one-click deploy" marketing talk. The install script does three things traditional one-click scripts don't:
1. Validates every input — a wrong API key is rejected on the spot, not three steps later 2. Idempotent — rerunning it breaks nothing and doesn't duplicate cron jobs 3. Environment-adaptive — if the VPS already has a reverse proxy (Hostinger/Coolify), it switches to a different publish mode instead of blindly spinning up another Caddy
Together, these three things mean non-technical users can install it. That's the long-tail problem open-source CRMs have never solved.
The AI Agent Isn't a Gimmick—It's the Core
Most CRMs' "AI features" are a chat box where you ask "what were this month's sales?" DeskcommCRM's AI agent does something else entirely: answer customers, qualify leads, close deals.
Concretely, the agent does three things on WhatsApp:
- Reception: when a customer messages in, the agent picks it up, understands intent, and judges whether it's a valid lead
- Qualification: per preset rules (QUANDO/SE/ENTÃO), it scores, tags, and routes leads into queues
- Selling: low-value leads get automated replies; high-value leads are escalated to humans
- Next.js 16: latest version, App Router + Server Actions, frontend and backend unified
- Supabase: Postgres + Auth + Storage, eliminating three services
- WAHA (WhatsApp HTTP API): open-source WhatsApp gateway, avoiding Meta's official API (expensive, slow approval)
- TypeScript strict: not optional, mandatory
This isn't an "AI assistant"—it's an "AI salesperson." The difference: an assistant answers your questions; a salesperson does the work for you.
MCP-Ready: Let Claude Code Install Your CRM
The most interesting part of the README:
> Drop the hostgator-setup-kit/ folder into a Claude Code chat running inside the VPS and say "install DeskcommCRM for me."
Hand the install kit to Claude Code, let it read the installation guide in CLAUDE.md, and it walks you through deployment. This isn't an "AI feature"—it's using AI agents as ops staff.
The project also ships a .agents/skills/deskcomm-instalar/ directory that Claude Code, Codex, Cursor, OpenCode, and Antigravity auto-load when opening the project folder. This is "skill-driven deployment"—instead of writing prompts for AI, you write skills for AI.
Why WhatsApp
DeskcommCRM was written by a Brazilian, and Brazil is WhatsApp's biggest market. In Brazil, small businesses don't use email, Slack, or DingTalk—all business happens on WhatsApp. So "WhatsApp-native CRM" isn't a feature in Brazil; it's a survival requirement.
But the positioning works globally too: India, Indonesia, Mexico, Nigeria—SMBs across all emerging markets do business on WhatsApp. DeskcommCRM isn't aiming at Silicon Valley's SaaS market; it's targeting WhatsApp sales scenarios for 500 million SMBs worldwide.
Tech Stack Judgment Calls
What's Worth Watching
DeskcommCRM isn't the most technically innovative project, but it gets several things right:
1. Puts the AI agent in the right place — not a chat box, but the execution layer of the sales workflow 2. Minimizes deployment friction — installable by non-technical users, the long-tail pain of open-source CRMs 3. Treats skills as the deployment interface — Claude Code can install it just by reading skill files, an early form of "AI ops" 4. Picked the right market — WhatsApp sales is a hard requirement in emerging markets, not a nice-to-have
505 stars isn't a lot, but the project's significance isn't in the star count. It proves one thing: the real use case for AI agents isn't "smarter chat boxes"—it's "process executors that do work for people." Sales CRM is the first vertical being rewritten by AI agents, and DeskcommCRM is a concrete sample of that trend.
---
GitHub: https://github.com/melgarafael/DeskcommCRM