DeskcommCRM is a free, MIT-licensed, self-hosted CRM that runs AI sales agents natively on WhatsApp, positioning itself against paid platforms like Kommo and Intercom that charge per seat and cap how many AI agents you can run.
A small business selling through WhatsApp, whether it is a clinic booking appointments, a real estate team, or an online store, usually ends up paying for two things at once: a CRM to hold the pipeline, and a metered AI layer bolted on top of it to keep leads from going cold overnight. On Kommo's pricing page, the Advanced plan runs $35 per user per month and caps a team at three AI agents. Move up to the Pro plan at $45 per user per month and the cap rises to 50, but it is still billed per seat, every month, indefinitely. DeskcommCRM, an open-source CRM that has climbed GitHub's trending list, runs its AI agents on a server you pay for once, under an MIT license, with no seat fee and no cap written into the pricing tier.
What DeskcommCRM actually does
DeskcommCRM describes itself on GitHub as an open-source "AI sales operating system" for WhatsApp, and names its target directly: it calls itself the open alternative to Kommo, Octadesk, and Intercom. The core idea is a CRM where AI agents sit inside the pipeline rather than beside it. Each tenant gets its own agent with retrieval over its own knowledge base, so the agent can answer questions, qualify a lead, move a deal through the funnel, and hand off to a human teammate when a conversation needs judgment the model should not make alone, according to the repository's README.
WhatsApp connects two ways: by scanning a QR code through the WAHA engine, with built-in throttling and randomized timing meant to reduce the chance of a number getting flagged, or through Meta's official Cloud API with approved message templates for teams that want the sanctioned channel. The CRM itself is multi-tenant, with the pipeline vocabulary configurable per business type, so a "lead" becomes a "patient" for a clinic or a "buyer" for a real estate shop without touching code. A follow-up system re-engages conversations that have gone cold on an adaptive schedule, and a webhook layer lets a business wire in leads from landing pages, forms, or tools like Zapier without custom integration work.
Licensing and release history back up the production claim rather than just asserting it. The LICENSE file is a standard MIT license with no added restriction, meaning commercial use, modification, and redistribution are explicitly permitted. The releases page shows 27 tagged releases, with v1.18.1 as the current version, rather than a project that only exists on an unstable main branch. As of this writing, the repository shows 1.3 thousand stars, 482 forks, and 43 contributors.
Who this is for
This fits a founder or a small marketing or sales-ops team already running its pipeline through WhatsApp and comfortable owning a server, or willing to pay a developer or agency to own it for them. The project ships a scripted, largely automated installer built with a hosting partner, aimed at getting a working instance onto a VPS without the person running it needing to install Node, a package manager, or compile anything by hand. That lowers the bar from "hire an engineer" to "follow a script," but it does not remove the VPS from the picture.
The honest caveats
Self-hosting is not free, even when the software has no license fee. The project's own documentation recommends a VPS with at least 4 GB of RAM, a Supabase account for the database, and an API key from an AI provider such as OpenRouter, Anthropic, or OpenAI, each of which bills separately for usage. Someone still has to point a domain at the server, keep backups current since the free tier of the recommended database does not back itself up automatically, and decide whether to accept optional error telemetry. Support is community-based and explicitly offered "as is," with no service-level agreement; the maintainers state plainly that whoever runs the instance is the one responsible for keeping it online and is the data controller for any personal information it processes.
There is also a real feature gap worth naming honestly. Intercom sells its Fin AI agent on a usage basis, at $0.99 per resolved outcome, layered on top of seat-based plans, plus optional add-ons like a $99-a-month analytics package and per-agent Copilot pricing. That usage-based model comes with a vendor's infrastructure, uptime commitments, and support staff behind it. DeskcommCRM's WhatsApp QR-code connection path is also explicitly built around anti-ban throttling for unofficial numbers, a workaround that exists precisely because that connection method is not the officially sanctioned one, unlike the Cloud API path it also supports.
A quieter observation
The most interesting number in Kommo's pricing page is not the monthly fee. It is the AI agent cap tied to the seat count: three agents on one tier, fifty on the next. That is a company deciding how much of its own AI capability to ration by price. DeskcommCRM's answer to that design choice is not that AI agents should be free of cost entirely, since a server and an API key still cost money, but that the number of agents you can run should not be a line item someone else controls.