A freelancer is the whole company: sales, delivery, support and billing, all from the same phone. Clients write on WhatsApp because that is where they already are, and every message pulls you out of the work you are being paid for. WhatsApp automation for freelancers is not about a bot that answers "hello, how can I help". It is about not losing the thread across ten clients while you are deep in a task.
WACrew is an open-source tool built for exactly that: one person, many client chats, and an AI that helps without speaking like a call center.
The problem it solves
- Context switching. You stop coding or designing to answer "did you see my message?" and lose twenty minutes getting back into the task.
- The forgotten client. The chat that arrived at 8 a.m. sinks under newer ones and nobody answers until the client complains.
- Voice notes and screenshots. Half of the brief comes as audio you have to stop and listen to.
- Generic auto-replies. The away message of the Business app tells the client you are not there, which is the opposite of what you want.
How WACrew handles it
One queue, sorted by who waited longest. Every group and direct chat lands in a single panel. The client who has been waiting since the morning is at the top, not the one who wrote last.
Voice notes become text. Audio is transcribed on arrival, so you read the brief in five seconds and search it later.
Replies in your voice. The AI drafts answers using your own writing measured from the messages you typed: how long your messages are, how you punctuate, how you open a chat. The gerar_voz tool builds that profile from your history, so the client reads something that sounds like you, not like a template.
You approve, or leave it on auto. Each chat starts in confirm mode: you see the text and press send. For the clients where the routine answers are always the same, you can turn on auto for that chat only.
Delivery at a human pace. Messages leave with a read receipt and a typing indicator, with the delay a person would take. If the client writes again before it goes out, the delivery stops and the agent rereads the chat.
Work the chats from Claude Code
If you already spend the day in a terminal with a coding agent, the WhatsApp MCP server lets that agent read what the client asked, check the screenshot of the bug and draft the reply, without you opening the phone:
claude mcp add wacrew -e WAC_CLOUD_URL=https://your-panel.example.com \
-e WAC_CLOUD_USER=admin -e WAC_CLOUD_PASS=change-me -- npx -y wacrew-mcp
Then ask things like "who is waiting for an answer?" or "what did the client ask today in the project group?". The full walkthrough is in How to connect WhatsApp to Claude Code with MCP.
Install
The collector has to stay online to capture messages, so it runs on a small server:
git clone https://github.com/sumioshi/wacrew && cd wacrew
PANEL_USER=admin PANEL_PASS=change-me docker compose up -d
Open /config, scan the QR code with WhatsApp (Linked devices), and the chats start to fill the panel. There is also a one-click Railway template and a prebuilt image; see the self-hosted inbox page.
What it costs
The code is free (Apache 2.0). You pay for the server, a few dollars a month on a small VPS, and for the AI provider you choose with your own key. Local transcription on a Mac runs without any key.
Honest limits
- Not the official API. WACrew links to your number as an extra device, like WhatsApp Web. It is for people who already talk to you. Cold outreach in volume is the fastest way to get a number banned.
- You run it. It is self-hosted: there is no account to sign up for, and backups are yours.
- The AI does not decide for you. Prices, deadlines and scope stay with you. In auto mode the agent answers what is already decided and hands the rest back.
FAQ
Does it work with my personal WhatsApp? Yes. It links as a device to a regular or Business number. Many freelancers use the same number for friends and clients, and the triage keeps personal chats out of the queue.
Will clients know an AI is involved? Only if the text gives it away, and the review before sending blocks the usual tells. Whether to tell your clients is your call.
Can I use it with Codex or Cursor instead of Claude Code? Yes. The MCP server works with any MCP client; see the WhatsApp MCP guides.
How is this different from a WhatsApp CRM? A CRM organizes contacts. WACrew reads the conversations, tells you who is waiting and helps you answer. If you want the organization side, read WhatsApp CRM for freelancers and agencies.



