How it answers your WeChat

One round, step by step

1. It reads the chat listAbout every 10 seconds, give or take 25%, it reads WeChat's chat list for chats with something new, even with WeChat behind your other windows, so it doesn't interrupt what you're doing. Muted chats and official accounts are skipped; a group you don't follow is opened only when you've been @-mentioned.
2. It waits for you to step away before opening a chatSomething new in the chat that's already open is read right there. Another chat has to be opened with WeChat in front, so it waits until the keyboard and mouse have been idle for 60 seconds (adjustable), then hands your window back. While you're in WeChat, it does nothing.
3. It sees who said whatIt checks whether the chat is one-to-one or a group, and reads a screenshot of the message list to tell your messages from theirs. Images, stickers and recall notices are ignored.
4. It decides what to answerOnly messages that arrived after your own last message count, at most the latest 5. They go to the model you chose; in a group, up to 20 earlier messages go along as context so the model can tell whether you're the one being asked.
5. It replies, signedIn dry run the reply is only a draft under Drafts in the menu. With a license activated and the menu switched from Dry run to Sending, it waits 2–6 seconds, types the reply, presses Return and checks the message has appeared. Every reply carries an AI signature.
6. It stays within limitsAt most 20 replies per chat per hour and 60 per hour overall. All of these are in Settings → Replies.

Your own account, WeChat left untouched

It works the copy of WeChat for Mac that's already logged in on this Mac. It reads the window through macOS accessibility, reads message bubbles from screenshots, and sends with real mouse clicks and key presses. It doesn't modify WeChat, inject code or use WeChat's private protocol. That's why it only works while the Mac is on and unlocked. WeChat's window can sit behind other windows, but not minimised; opening another chat, a profile card or sending a file waits until you're away, then brings WeChat forward.

A WeChat update can change its window. Meow looks at WeChat's window, read only, at launch and whenever WeChat's version changes; if it can't read it, sending pauses instead of failing chat by chat. The result is in Settings → General → WeChat, with a Check Now button. WeChat can be in Simplified Chinese or English.

This kind of UI automation isn't allowed by WeChat's terms, and an account could be restricted. Read account safety before you turn sending on.

One-to-one chats

• It answers only what arrived after you left. Anything you've already answered yourself, it leaves alone.
• One to three sentences, in the sender's language. It speaks as your AI assistant and makes no promises for you: meetings, prices and decisions get passed on to you.
• It's told not to guess, and to say you'll confirm when it isn't sure. That's an instruction to the model, which can still get things wrong.
• Within a chat it remembers the last 30 minutes (up to 30 entries), in memory only.
• Messages that arrive while Meow isn't running are not answered later.

Groups: only what's meant for you

• Groups you follow (Settings → Groups → Followed groups): every new message goes to the model, which decides whether it's for you: an @-mention, your name, a question to you, or a thread you're part of. If not, it stays quiet.
• Other groups: opened only when someone @-mentions you or quotes your message. The switch is “Also look at other groups when @-mentioned”, on by default.
• After it replies in a group it keeps watching for 30 minutes, because the answer rarely @-mentions you again.
• Group replies are one or two sentences, written for everyone in the group to read. Groups get web search and the public folder only, never the private folder.
• Every check in a followed group is one model request, billed even when it stays quiet. A busy group is usually most of the bill.

Three sample conversations

These are scripted examples, not real chats: a one-to-one chat, a group you follow, and a group you don't.

Dry run first, then let it send

Out of the box it runs dry: it reads your chats and drafts replies but sends nothing. Every draft appears in full under Drafts in the menu. Drafts live in memory only: they're gone when Meow quits and never reach a log. Watch it for a few days, then decide whether to buy.

Sending needs a license. Activate your key, then click Dry run at the bottom of the menu so it reads Sending (it asks you to confirm). If the license ends or is revoked, the app drops back to dry run on its own. How to activate

The signature

Every reply ends with an AI signature on its own line. The default follows the app's language: “— Replied by AI assistant” in English, “— AI 助手代回” in Chinese. You can reword it in Settings → Replies but not remove it: cleared, it falls back to the default. Asked directly whether it's a person, it says it's an AI assistant answering for you. It doesn't say which company's model it runs on.

Files, remote commands and mail

• Files: the public folder (~/meow/Public) can be used for anyone; the private folder (~/meow/Private) only for contacts you trust and for you. It reads text, Word, RTF, HTML, PDF and images; PDFs and images up to 10 MB each. DeepSeek's endpoint doesn't take PDFs.
• Remote commands: tell it in File Transfer on your phone (Settings → You → Answer me in File Transfer), or message it from your other WeChat account, to send a file to a chat or pass a message on. Another account it recognises by the WeChat ID on the contact's profile card, never by the nickname.
• Contacts: Settings → You → Scan Contacts… reads your contacts list once; after that, ask “Who do I know in London?” and it knows. The list stays on this Mac, and only you can ask.
• Mail alerts: a Chrome extension reads the unread rows of your Gmail and Outlook web inboxes and sends the important ones to your other WeChat account as one-line alerts. In Gmail it can also draft a reply. It never sends one. Setup