codex / WeChat Bridge中文GitHub

From WeChat,keep using Codex

Connect personal WeChat to your own Codex.
Use Skills, run commands and read repositories, then continue on the Web.

Get connected See what it can do
Codex WeChat Bridge127.0.0.1:8787
Web session manager with conversation history, model selection and attachments (Chinese interface)
Actual WeChat conversation asking about the weather in Chengdu (Chinese)

Start in WeChat. Continue on the Web.
One Codex, with context intact.

Real conversations,
real tasks

Send a task in WeChat; Codex runs it on your deployment machine.
Use Skills, switch models, check the weather or read a repository.

Use your Skills

Name the installed amazing-images-template Skill in WeChat and choose template 1 to start a travel poster task. The screenshot shows template selection and example feedback.

Use the image template Skill to turn a photo into a travel poster with an extended line drawing.
Open full screenshot ↗

Actual WeChat conversations supplied by the project author. Screenshots are in Chinese. Skills and tools must be available in your Codex environment.

Use your Skills

Send tasks.
Get results.

From a photo to a project, from a question to ongoing collaboration.
Work with your own environment through a familiar conversation.

Codex reads a GitHub repository and returns its purpose and setup steps in WeChat (Chinese)
WeChat · Repository setup

Bring your Skills into WeChat

Name a Skill installed in your deployment and start a task with text and images. The example calls an image template Skill and selects a travel poster style.

Receive results in the conversation

Codex can return local images, videos and files as native WeChat messages. Audio is sent as a file, so you can open the results from your phone.

Follow longer tasks as they run

Progress is shown by default and can be collapsed on the Web. Send /stop when you need to interrupt the current task.

Keep the context with your session

Start with /new or switch to history with /resume. WeChat and the Web share one Codex thread, with model and reasoning settings per session.

Continue a conversation

Review Markdown history, add attachments and continue the same Codex thread from WeChat in the Web interface.

Separate spaces for each account

Independent connection status, contact authorization and managed sessions. Scan again for an existing account while retaining local notes and history.

Set your execution boundaries

Configure workspaces, allowed directories and sandbox policy. The Web supports login and update checks; npm installations can update and restart from the page.

Your choice of
where to run

Your own computer, or an online remote server.
Tasks run on that machine using its Codex environment.

Let your computer handle WeChat tasks

Run the bridge on your Mac, Windows or Linux computer. Codex uses that computer’s workspace; the management page is local by default.

  • Requires Node.js 22+ and an installed, signed-in Codex CLI
  • Default management page: http://127.0.0.1:8787
  • Keep the computer and bridge running while processing tasks
Install and start
npm install -g @agent-dev-net/weixin-bridge@latest
codex-weixin

The management page opens in your browser after startup.

Read the full deployment and environment guide

Send your next task
from WeChat

From environment setup to your first task.
Follow these steps to start your first bridged conversation.

Install and sign in to Codex CLI first

npm install -g @openai/codex
codex
Requires Node.js 22+. Follow the sign-in prompts on first launch.
WeChat account page with QR sign-in and contact authorization (Chinese interface)

  1. Install and start the bridge

    After signing in to Codex, install the npm package using the deployment instructions above. Run codex-weixin to open the management page.

    View deployment commands

  2. Set your workspace

    In Settings, confirm the default working directory and allowed directories. Choose a model, sandbox and approval policy as needed, then save.

  3. Scan with WeChat

    Choose Add WeChat, scan the QR code and confirm sign-in. Repeat to connect multiple personal WeChat accounts.

  4. Authorize a contact

    Send a message to the connected account from WeChat. Return to WeChat Accounts and choose Allow access for the new request.

  5. Send your first task

    Try: “Read this project’s structure and explain each directory.” Open Sessions on the Web to continue the conversation.

Switch sessions
with a command

/newStart a new managed session
/resumeView history and continue
/modelView or switch the model
/stopInterrupt the running task
See all WeChat commands

Before you
get connected

Will tasks run when my computer is off?

The bridge and Codex run on the deployment machine. Tasks cannot run when it is powered off, asleep or the service is stopped. Use an online remote server for continuous availability.

What do I need to get started?

Node.js 22+, an installed and signed-in Codex CLI, and personal WeChat for QR sign-in. The bridge uses the deployment machine’s Codex login and workspace settings.

Can I continue between my phone and the Web?

Yes. WeChat and the Web share Codex threads managed by this bridge. It only manages sessions created and used by this service, without taking over all history from other terminals.

How large can attachments be?

Each WeChat attachment can be up to 100 MiB. The Web accepts up to 10 files per upload, totaling at most 100 MiB. Files are isolated by WeChat account and session.

What does remote deployment require?

Install Node.js, Codex CLI and the bridge, then set the listening address, allowed hosts and Web credentials. Access through an actual IP or domain. For public access, use an HTTPS reverse proxy and restrict access to the service port.

How do I control access and execution?

WeChat contacts require explicit authorization. Enable Web login through environment variables and configure working directories, allowed directories, sandbox and approval policy in settings. The bridge currently has no interactive approval interface; approval requests are rejected.