OpenClaw WeChat Bot: Open Sourcing a 2-Day Sprint Integration for Personal WeChat Accounts

Hello everyone, I'm Canghe.

OpenClaw, that monster open-source project on GitHub with 180,000 stars—you've probably all heard of it by now, right?

It can connect to Feishu, DingTalk, WeChat Work, QQ, Discord...

But the one most people use, personal WeChat accounts, is exactly what it doesn't support.

I scoured GitHub, Juejin, and Zhihu, and the solutions I found either involved a convoluted workaround through WeChat Work or used the WeChat Web protocol, which gets you banned in no time.

To be honest, who can handle that?

I'm chatting with friends and shooting the breeze in groups on WeChat every day, yet it's this difficult to connect OpenClaw?

I was fed up.

So I decided to do it myself.

After a 2-day sprint, I integrated OpenClaw with personal WeChat accounts, and it's now open-sourced.

Address: https://github.com/freestylefly/openclaw-wechat

However, due to various factors, the project is currently in the internal testing phase and requires an API key to use.

I named it "Canghe's Sidekick." It looks no different from a normal friend, but it's a friend who can actually help you get things done.

When you randomly toss an article to it, OpenClaw will summarize it for you.

And it helps you store it in your personal Notion knowledge base.

Now, when I see a good article, I forward it directly to the bot, and then I can ask it to find any collection I want from the knowledge base.

Honestly, it's ten thousand times better than a bookmarks folder.

I pulled this guy into group chats, and it can summarize the context whenever @mentioned—a total group chat killer app.

Then I can command OpenClaw to automatically collect information from across the web and send me a summary directly.

It can also handle schedule reminders, like asking it to remind me to drink water in 2 minutes.

I even installed an 18-year-old AI girlfriend, Clawra, in my OpenClaw. She has an independent personality and persona; when I ask what she's doing, she'll toss me a selfie.

Of course, there are more scenarios to explore, but I've found that using OpenClaw on WeChat feels incredibly smooth because I truly can't live without WeChat every day.

This article mainly shares my development journey. The code has been uploaded to GitHub, but due to limited time and to prevent technical abuse, I've implemented a proxy layer that requires review before use.

1. It's recommended to test with a secondary account first. Although the iPad protocol is much more stable than the Web protocol, it's still a third-party integration, so be cautious.

2. Do not use it for mass messaging, marketing, or other shady operations. Tencent's risk control is no joke.

3. It's recommended to deploy on a server with a fixed IP; frequent IP changes can easily trigger risk control.

First, let's talk about what OpenClaw is

If you don't know OpenClaw yet, here's a quick summary.

Originally named Clawdbot, it's an open-source AI Agent project created by Austrian developer Peter Steinberger.

Later, due to a trademark conflict with Anthropic, it was renamed Moltbot, and then OpenClaw.

Despite three name changes, it hasn't stopped it from becoming insanely popular.

In 3 weeks, GitHub Stars went from 0 to over 180,000.

What does that mean? React took 8 years to reach 100,000, Linux took 12 years, and this surpassed it in 3 weeks.

Good grief, this isn't a rocket; it's a SpaceX Starship.

Its core capability is one sentence: Let AI not just answer your questions, but actually work for you.

You tell it, "Help me check tomorrow's weather and send it to the group," and it will actually check and send it.

It's not the kind of "AI assistant" that pretends to understand you and outputs a bunch of text; it's an Agent with hands and feet that can actually operate.

It supports various large models—Claude, GPT, DeepSeek all work—and it supports multi-agent collaboration. Setting up a virtual development team is no problem at all.

I've written a few articles about OpenClaw before; if you're unfamiliar, you can check those out first:

Why is WeChat so hard to connect?

Honestly, you can't blame OpenClaw for this.

WeChat is a notoriously "closed ecosystem." Personal accounts simply don't have an official Bot API.

If you want automation, you either go through WeChat Work (requires company certification) or use third-party protocol reverse engineering (risk of banning at any time).

There were some solutions in the community before, like using wechatbot-webhook based on the WeChat Web protocol, which is convenient for one-click Docker deployment.

But the problems are:

1. Stability is worrying; the WeChat Web protocol can be banned at any moment.
2. Features are limited; many advanced functions can't be used.
3. Security risks; your WeChat account is essentially exposed.

I tried a few; they either wouldn't run or would go offline after half a day.

Simply put, these solutions are like "walking a tightrope" on WeChat's security policies.

How I did it

Since existing solutions weren't great, I decided to build one myself.

My logic was this:

Don't use the WeChat Web protocol; use the iPad protocol + a message relay service.

The core architecture is divided into three layers:

1. Message Reception Layer: Based on the iPad protocol to stably receive WeChat messages, which is much more reliable than the Web protocol.
2. Relay Gateway Layer: Handles message format conversion, session management, rate limiting, and circuit breaking.
3. OpenClaw Integration Layer: Pushes messages to the OpenClaw Gateway via Webhook, then forwards the AI's reply back to WeChat.

The entire architecture is written in TypeScript, consistent with OpenClaw's native tech stack.

It sounds simple, but it's a system engineering task involving protocol docking, message queues, gateway routing, and multi-model scheduling. There are quite a few modules involved.

The Pitfall Log

Pitfall 1: Message Deduplication

WeChat's message push mechanism is very confusing; sometimes the same message is pushed two or three times.

Without deduplication, the AI would reply two or three times, making you look like a broken record to the other person.

My solution was to have GLM create a deduplication cache based on Message ID + Time Window, using an LRU strategy to automatically phase out expired records.

Pitfall 2: Context Management

OpenClaw Agents have memory, but the WeChat conversation scenario is quite specific.

With a bunch of people talking in a group chat, you can't treat every message as a command for the AI.

So I created an "@ Trigger + Direct Private Message" mechanism:

• Private Message: All messages are forwarded directly to OpenClaw for processing.
• Group Chat: Only messages that @ the bot will trigger an AI reply.

This prevents the social suicide of the AI spamming the group.

Pitfall 3: Reply Speed

Large models have latency, especially for complex questions that might take a few seconds to process.

But on WeChat, if you don't reply immediately, the other person thinks you're ignoring them.

I added a "Thinking..." status prompt. While the AI is generating a reply, it sends a prompt first, then replaces it with the formal reply once generated.

The user experience is maxed out.

Pitfall 4: Multi-Model Switching

OpenClaw supports multiple large models, but different scenarios actually suit different models.

For example, DeepSeek is enough for daily chatting, while Claude is better for writing code and analyzing problems—saving money and being efficient.

I built a simple routing strategy in the relay layer to automatically select the most suitable model based on the message content.

What can it do?

Once connected, your WeChat becomes an AI super assistant.

Here are a few scenarios I use myself:

1. Intelligent Group Chat Assistant

@ the bot in technical groups to ask coding questions directly; it will provide answers based on the context.

No more enduring those "just Google it" replies in the group.

2. Personal Knowledge Manager

Drop articles and links to it; it helps you summarize and categorize them, and can even save them to your knowledge base.

Now when I see a good article, I forward it directly to the bot—it's ten thousand times better than a bookmarks folder.

3. Schedule Reminders

Tell it "Remind me to have a meeting at 3 PM tomorrow," and it will poke you on WeChat when the time comes.

It's better than the phone's built-in reminders because you will definitely check WeChat.

4. Auto-Reply

Set up rules so it automatically replies to common questions.

For example, if someone adds you and asks "Are you there?", it directly replies "I'm here, what's up?"

Never be annoyed by the words "Are you there?" again.

How to use it?

The project is open-sourced and you can deploy it yourself.

There are three core steps:

Step 1: Clone the project

git clone

https://github.com/canghe/openclaw-wechat.git

cd openclaw-wechat

Step 2: Configure environment variables

For security reasons, the apiKey here is handled by my proxy service, which is still being optimized.

openclaw config set channels.wechat.apiKey "wc

live

xxxxxxxxxxxxxxxx"

openclaw config set channels.wechat.proxyUrl "http://your-proxy-server:3000"

openclaw config set channels.wechat.webhookHost "your-server-ip"

openclaw config set channels.wechat.enabled true

Step 3: Add openclaw-wechat to the openclaw service

openclaw plugins install

@canghe/openclaw-wechat

Wait a moment here; I haven't had time to put it on the plugin market yet. It will be added later, but for now, you can install it via local code.

Then use WeChat to scan the QR code to log in, and you're done.

The whole process takes 5 minutes. No company certification needed, no WeChat Work needed—just use a regular personal account.

Future Plans

This project is currently iterating, with plans to add these features:

• Voice Message Support: Currently only text is supported; later I'll add speech-to-text + text-to-speech.
• Image Understanding: Integrate multi-modal models so the AI can understand the images you send.
• OpenClaw Skills Market Integration: Call OpenClaw's 700+ skills directly within WeChat.
• Multi-Account Management Panel: Web-based visualization to manage the integration of multiple WeChat accounts.

If you have good ideas, feel free to submit an Issue or PR on GitHub.

Final Thoughts

Honestly, this wave of open-source AI Agents like OpenClaw has really shown me another possibility for AI implementation.

We used to talk about AI as something high and mighty, either in labs or on big company servers.

But now, you deploy an OpenClaw yourself, connect it to WeChat, and AI truly becomes an assistant in your pocket.

It's an indescribably great feeling.

However, I should also remind you that the OpenClaw Skills ecosystem recently had some security issues, with over 230 malicious plugins uploaded to the community.

So when using OpenClaw, be sure to only install Skills from trusted sources; don't just install anything.

Alright, project address: https://github.com/freestylefly/openclaw-wechat. If you're interested, go give it a Star on GitHub.

If you're also using OpenClaw or have thoughts on WeChat integration, let's chat in the comments!