It is a WhatsApp sales system, not just a contact list
DeskcommCRM comes from a clear Brazilian context: its main README and installation guidance are in Portuguese, with English and Spanish translations. The project positions version 1.18.1 as an open alternative to Kommo, Octadesk, and Intercom, built around WhatsApp sales rather than generic customer records. Its MIT license also gives agencies and internal teams room to adapt it.
The product combines a self-hosted CRM, WhatsApp connectivity, sales automation, and AI agents intended to answer, qualify, and sell. Its documented stack includes Next.js 16, strict TypeScript, and Supabase for Postgres, authentication, and storage. Operators can connect OpenRouter, Anthropic, or OpenAI while keeping the application and customer data on infrastructure they control.
What happened when we ran it
We cloned commit 30bd6f5 into a fresh, unprivileged Debian container with 3 CPUs and 8 GB RAM. Installation succeeded in 45 seconds through pnpm, bringing in 923 packages and using 1,251 MB on disk. The production build completed successfully in 195 seconds. That is a credible first run for 3,951 files, about 504,872 source lines, and an 86.2 MB checkout.
Testing is the unresolved part. The repository had a tests directory and 7 CI workflow files, but no tests script or target that our harness could execute, so tests were skipped. The successful build proves the build path worked in our container, not that message delivery, authentication, migrations, automation queues, or rollback behaved correctly.
The installer addresses real production chores
The strongest part of the pitch is the operational detail. The setup kit says it generates technical secrets, installs Postgres extensions, applies the baseline schema, creates the first administrator, starts the stack with HTTPS, installs the automation cron, and enables the update agent. The README recommends 4 GB RAM and provides both an idempotent interactive path and a prepared non-interactive mode.
Proxy handling reflects problems self-hosters actually encounter. The installer checks whether ports 80 and 443 are already occupied and adapts instead of blindly starting Caddy. For a host-network proxy configuration, it asks the operator rather than guessing. Updates can back up the database, move through visible stages, and roll back, while an agent checks requests every 5 minutes.
Onboarding still crosses several services. You need 1 Docker-capable VPS, a domain with an A record, Supabase credentials and a session-pooler string, an AI provider key, and a WhatsApp number connected by QR code or Meta's official channel. Optional 2-factor authentication uses Google Authenticator or Authy. This is manageable for an operator, not a no-ops experience.
The rough edges are operational and evidentiary
The dependency footprint is large: our installation placed 923 packages across 1,251 MB before the build. That is not automatically excessive for a full CRM, but it widens the maintenance surface. Docker and Compose files help packaging, yet operators should still scan images, control upgrades, protect environment files, and rehearse database restoration rather than assuming scripts cover every local change.
The missing executable test target is the clearest caution. A production build is not a regression suite for customer conversations. With 89 open issues, evaluators should inspect reports relevant to their exact WhatsApp channel and hosting pattern. The supplied evidence does not show response or closure times, so the issue count alone cannot establish either healthy engagement or neglect.
Language is another consideration. English documentation exists, making evaluation possible outside Brazil, but the primary material is Portuguese-first. A 3-language README set is welcome, although international teams should verify the deeper runbook and interface language. The promoted 1-command curl route is convenient; on a customer-data server, cloning and inspecting the local setup kit first is the more defensible practice.
Community activity is promising, not conclusive
The latest release, v1.18.1, and the last push both landed on 2026-09-11, which is strong evidence of current development. Seven CI workflow files show attention to automation, and the repository appeared in the supplied GitHub Trending item. None of that tells us how quickly maintainers review security reports, support upgrades, or resolve channel-specific breakage, so a pilot still needs its own acceptance checks.
The project has 1,089 stars and 89 open issues, enough visibility to suggest real interest but not enough evidence to infer a dependable support organization. The README is unusually candid about update-agent delays, rollback, proxy conflicts, and optional 2-factor authentication. That practical documentation raises confidence, while the absent runnable test result prevents a higher maturity judgment from our review.
Where it fits in a real stack
DeskcommCRM sits at the application layer, with Supabase holding data and identity, WhatsApp providing the customer channel, and OpenRouter, Anthropic, or OpenAI supplying AI capability. A reverse proxy and HTTPS expose it, while cron drives automation and the 5-minute update polling cycle. The recommended 4 GB VPS must be treated as production infrastructure, with monitoring, backups, access controls, and recovery ownership.
Choose it when WhatsApp is central to revenue work and your team can operate that 6-part chain of app, database, messaging, AI provider, proxy, and scheduled jobs. Avoid it when you want vendor-owned uptime or need a verified automated test baseline today. The 45-second install and 195-second build make a controlled pilot reasonable, but live adoption should follow channel, restore, migration, and rollback exercises.