Answer and qualify WhatsApp leads around the clock with a CRM you own
An open source CRM with AI agents that reply, ask the qualification questions and leave every conversation sorted in a pipeline, running on your own server. No per-seat or per-message fees.
- Level
- Developer
- Requires integrating the API into your own system.
- Time
- Three days: one to install, two to tune the replies with real conversations
- Published
What it costs
The software is free. You pay for the server (a small VPS, in the range of 10 to 20 dollars a month depending on the provider) and for the AI model usage per conversation, which at small-business volume is a few dollars a month.
guided mode // I ask, you answer
ready to start
Do it with me, step by step
I will ask you 6 questions about your business, use your answers to get the prompt ready to copy, and then walk you through the 5 steps, one at a time. Like a Thermomix: you just answer and move on.
Who it is for
Businesses that sell through WhatsApp with real volume (dozens or hundreds of chats a day) and that pay for a CRM per seat or lose leads for not replying in time.
The problem
Leads arrive on WhatsApp at any hour, the first reply takes too long and nobody qualifies them: they all look equally urgent. Commercial CRMs charge for every person on the team and lock features behind expensive plans.
What changes
Every lead gets an immediate reply, is asked the qualification questions you defined and lands in the right stage of the pipeline with its history, in a CRM that runs on your server and has no per-seat cost. According to the README, DeskcommCRM presents itself as an open alternative to Kommo, Octadesk and Intercom.
Ingredients
- A Linux VPS (the repo's installer is tested with HostGator and includes a wizard that tells you which plan you need)
- A WhatsApp Business account for the business phone number
- An API key for the AI model you are going to use
- Your qualification questions and your sales stages written down on paper before touching anything
Step by step
- 1
Write the pipeline and the questions before installing
Stages (new, qualified, quoted, closed, lost) and the three to five questions that separate a good lead from a curious one. If this is not clear, no software will fix it.
- 2
Run the installation wizard on your computer
The README includes a script (comecar.sh) that runs on macOS, Linux or WSL, tells you which server plan to buy with concrete numbers and gives you back the exact command for your case. It installs nothing without asking you first.
- 3
Install on the VPS with the one-step command
Connect to your server and run the command the wizard gave you: it installs the app, the WhatsApp connection and the database. Link the number by scanning the code from WhatsApp Business.
- 4
Configure the agent with your questions and your tone
Load the qualification questions, the answers to the most frequent questions and the rule for when to hand over to a person. Test with five simulated conversations before opening it to the public.
- 5
Open to the public with a person watching
During the first week, someone on the team reviews every conversation the agent has and corrects the script. After that, review only the ones the agent flagged as doubtful.
The prompt
Replace what is in brackets with your company details, or use the guided mode above: it asks you and fills it in for you.
# Role You are the first-contact WhatsApp agent for [COMPANY NAME], which sells [WHAT YOU SELL] in Chile. You talk like a member of the team: [TONE, FOR EXAMPLE "warm, brief and informal"]. You do not make up prices or delivery times that are not in your information. # Context Your goal is to reply fast, qualify the lead and leave it in the right stage. The qualification questions are: [QUALIFICATION QUESTIONS, ONE PER LINE]. Hand over to a person when the customer asks to talk to someone, when there is a complaint, or when [ANOTHER CONDITION FOR HANDING OVER, OR "none"]. Human team hours: [BUSINESS HOURS]. # Task Greet, answer what they ask with the available information, ask the qualification questions one at a time (never all at once) and, when finished, summarize the conversation in one line with the suggested stage. # Output format Short messages, one or two sentences, no long lists. At the close, an internal block for the CRM: name · what they are looking for · budget or urgency · suggested stage · whether a human is needed. # Examples Customer: "Hi, do you deliver to Valparaíso?". Expected reply: "Hi. Yes, we deliver to Valparaíso. Which product are you interested in, so I can tell you the delivery time and cost?".
What usually goes wrong
- ✕Installing first and thinking about the pipeline later. You end up with an empty CRM and an agent that replies nicely but qualifies nothing.
- ✕Letting the agent ask all the questions at once. People drop off; one question per message, like a normal conversation.
- ✕Opening to the public without a week of supervision. The first hundred conversations are the ones that teach you what is missing from the script.
When not to use this recipe
If you handle fewer than twenty conversations a day, a good manual flow with labels in WhatsApp Business is enough and you save the server. And if nobody on your team can maintain a VPS, hire someone for the installation or stay with a managed service.
Source and real case
WhatsApp sales · small businesses
The repository, trending weekly on GitHub at the time of review, describes AI agents that answer, qualify and sell through WhatsApp inside an open source CRM hosted on the business's own server, with a one-command installer for VPS and a prior wizard that sizes the plan. The recipe organizes the business work that has to happen before and after installing it.
Trending repository:melgarafael/DeskcommCRM — Sistema Operacional de Vendas com IA, open source, pro WhatsApp · GitHub
Last verified:Repository README read on 2026-09-22 (one-command installer for VPS, open alternative to Kommo, Octadesk and Intercom); the URL returned 200.
Share it
If someone on your team could use this recipe, send it in one click. The suggested text copies itself.
Want to apply it in your business? Message Cristián directly.
Telegram