DeskcommCRM

DeskcommCRM is the self-hosted answer to seat-priced WhatsApp CRMs: instead of renting a conversation platform per user, you install the whole sales desk on a server you own.

Read on deskcomm.com.br on 27 September 2026, the product describes itself as a living system, and the rules are unusually concrete.

An AI agent reads the conversation before it replies, runs seven guardrails before every outbound message, and moves the lead stage itself.

Each guardrail evaluation becomes a durable, exportable record - including the ones that blocked a send - so a manager can audit what an agent said and why, which is exactly the gap that makes most WhatsApp bots unusable in a sales process.

When the agent should not act alone it hands the request to a human with a summary of what happened, what was agreed, which objections came up and what the next step is, rather than dumping the raw thread.

A Radar view classifies every open request as critical, at risk or in flight based on time without interaction and what is still pending, and an activity timeline records every meaningful change - so follow-up is driven by the current state of the pipeline instead of last month report.\n\nThe engineering claims are testable because the code is public.

When we checked the GitHub API on 27 September 2026, github.com/melgarafael/DeskcommCRM reported 4,058 stars, 989 forks, an MIT licence, TypeScript as the main language, a repository created on 28 April 2026 and a last push the same day we looked.

The repository description calls it an open-source AI sales OS and an alternative to Kommo, Octadesk and Intercom, with WhatsApp handled through WAHA, MCP-ready endpoints, multi-tenancy and LGPD alignment.

The vendor site points each claim at a file: tenant isolation is tested on every change with 364 invariant tests in CI - one test creates two organisations and proves one sees zero rows of the other, with a control case proving the rows exist - the seven pre-send guardrails live in lib/agent-engine/guardrails/before-send.ts, and the upgrade path is exercised in CI rather than merely documented.

Installation is one command on a VPS: clone the repo, run install.sh from the hostgator-setup-kit, and the script checks dependencies, generates secrets, applies the database, creates the first admin, brings up the containers and wires the scheduled automation task, with HTTPS issued on first access.

The documented footprint is seven containers, roughly 150 MB per WhatsApp session and 4 GB of RAM recommended - the stack boots on 2 GB but the site says it runs at the edge.\n\nPricing is the part that will surprise most buyers.

Read on deskcomm.com.br pricing on 27 September 2026, the software is free under the MIT licence with no paid tier, no gated feature and no per-seat charge: what you pay for is the VPS and the AI keys you consume, and the vendor makes the argument that five seats or fifty cost the same server.

Note the discrepancy we found: topai.tools lists DeskComm as Freemium, while the official site states plainly that there is no paid tier - we go with the vendor page and flag the difference.

Agencies are explicitly welcomed: the MIT licence allows hosting for third parties and charging for it, and two .env variables change the name and logo across the interface.

The honest limitations are self-hosting and scope.

WhatsApp is the only conversation channel, there is no managed cloud or vendor SLA, documentation and community are Portuguese-first although the site carries English and Spanish versions, branding is per installation rather than per organisation, and colour or font changes require working in the design system.

For a team that already runs its own infrastructure and sells by WhatsApp, that trade is favourable; for a team that wants a support line, it is not.