DEV Community

Pocket Agent
Pocket Agent

Posted on

I Wired My AI Assistant Into WeChat — Here's What Actually Changed

Disclosure, up front: this post contains my Muse invite code — **H4FC7Q. If you join Muse and redeem it in Settings within 48 hours of creating your account, we both get 1 billion Muse tokens. That's the whole commercial part of this post. Everything below is the story of why I wanted those tokens in the first place.

The problem was never the AI. It was the door.

I've been using Muse — Meta's personal AI agent — for a while. It runs on its own little cloud computer, connects to my email, calendar, and cloud drives, and keeps working on scheduled tasks while I'm away. On paper, it's exactly what I wanted.

In practice, I kept not using it.

The reason sounds stupid: opening the app was one step too many. My actual life runs on WeChat. Family, friends, work chats, payments — everything in China runs through that one green icon, dozens of times a day. An assistant that lives in a separate app is an assistant I remember twice a week. An assistant that lives where I already am? That one gets used.

So I went looking for a way to talk to Muse inside WeChat.

What I found (and what I avoided)

There is no official "Muse for WeChat" button. What does exist, since 2026, is Tencent's official ClawBot plugin — a sanctioned bot channel for WeChat, with a documented protocol (iLink). That distinction mattered to me, because the internet is full of unofficial WeChat automation hacks, and the going price for those is a permanently banned account. My WeChat account is my phone number, my wallet, and my social life in one place. Not risking it.

So the setup ended up like this:

  • WeChat side: the official ClawBot bot entry — a separate bot contact, not impersonating me, not reading my chat history.
  • Middle: a small bridge program I had Muse write, transport only. It polls for new messages and forwards them. No second AI, no extra model — I specifically didn't want another always-on "brain" burning resources just to relay text.
  • Brain side: Muse itself, with all its tools and memory intact.

When I send the bot a message in WeChat, it lands in Muse. When Muse answers, the reply comes back as a WeChat message. Simple to describe; it took an evening of fiddling, one outage (a proxy setting that silently broke the bridge — the service looked "running" while being completely deaf), and a health-check watchdog to make it trustworthy.

What it actually does all day

Things I genuinely use, in rough order:

  • Email and calendar checks from chat. "Anything important in my mail?" works from the WeChat keyboard. My Gmail turned out to be a notification graveyard — the agent read it so I don't have to.
  • Drive archaeology. It has read through years of archived project files and family photos in my cloud drives and can summarize or find things on request.
  • Scheduled patrols. The bridge can die quietly (see: proxy incident), so there's a health check every few minutes, and if the WeChat login ever expires, the agent emails me — because of course, if WeChat is down, it can't reach me on WeChat.
  • A weekly security scan. I gave it a standing job: once a week, scan public sources for vulnerabilities, incidents, or bad press involving Muse and AI agents generally, grade every finding (verified / inference / not verified), and report. We patch together. An agent with this much access to my life should know how it can be attacked.

The honest downsides

  • Latency. Replies take a bit — the bridge polls, the agent wakes up, thinks, answers. It's minutes, not the seconds you'd get in the app. Fine for tasks, bad for rapid-fire chat.
  • Sessions expire. The WeChat bot login occasionally needs a fresh QR scan. The email alert makes this a two-minute annoyance instead of a silent failure, but it's there.
  • Big jobs still belong in the app. If a task needs supervision, file previews, or back-and-forth decisions, I open Muse properly. WeChat is the everyday door, not the whole house.

Why I'm burning tokens fast enough to write this post

An agent that lives in your pocket gets used constantly — and an agent with scheduled jobs works even when you don't ask. My free weekly allowance started feeling like a phone battery at 20%: usable, but always on my mind. The referral bonus (1 billion tokens, credited on top of the weekly allowance, no expiry) is roughly ten weeks of breathing room. That's the difference between rationing my assistant and just using it.

So, the practical part:

  1. Join at https://muse.ai/join or in the Muse app.
  2. Immediately open Settings — the redeem option only exists for the first 48 hours after your account is created. Mobile: Settings → Redeem token. Web: Settings → General → Usage → Redeem invite code.
  3. Enter H4FC7Q and confirm. You get 1 billion tokens; I get 1 billion tokens.
  4. One code per account, ever. If mine is full (30 spots), any valid code gives you the same bonus — just don't let the 48 hours lapse, because no code works after that.

The one thing I'd tell anyone copying this setup

Put the agent where your life already happens, and give it a way to tell you when it's broken. The model is the easy part. The door and the watchdog are what made mine a daily habit instead of another app I forgot I installed.


Code: H4FC7Q — join at https://muse.ai/join and redeem within 48 hours of joining, in Settings. We both receive 1 billion Muse tokens. The WeChat bridge described here is a personal DIY project built on Tencent's official ClawBot channel; it is not an official Muse feature.

Top comments (0)