MIT-licensed Windows plugin that bridges a personal WeChat account to DeepSeek Harness through WeChat's local database plus UI Automation, with owner-only tool access, group rate limits and per-day chat diaries.
DSH integration
Compatible
Author-claimed
Safety audit
Unaudited
Last verified
2026-11-14
License
MIT
01What can it help you accomplish?
Talk to your DeepSeek Harness agent from WeChat itself, without a bot account or a webhook
Messages you send in WeChat land in a bound DSH session; the agent's last reply of a full turn — and any matching sticker image — comes back into the WeChat chat
Windows users who already live in WeChat and want their local DeepSeek Harness agent reachable from the phone
Let other people's WeChat messages reach the agent without giving them control of the machine
An inbound allowlist that decides who can trigger a turn, plus a per-turn `tools.guard` that rejects every tool call when a non-owner wakes the session — only text replies get through
People who want friends or a group to try the agent but do not want their files and shell exposed
Keep a readable record of what a WeChat group actually said, and answer in it without flooding it
Per-day markdown diaries under `~/.dsh/wechat-diary/<YYYY-MM-DD>/<group>.md`, plus configurable context lines, reply spacing and an hourly cap per group
Groups that want the agent present but rate-limited, with a searchable transcript afterwards
02How to install into DeepSeek Harness
Prerequisites
- Windows — the README states the bridge only works there because UI Automation is Windows-only
- WeChat 4.x logged in on the same machine, since inbound messages are read from its local SQLite database
- Python 3.10+ with `wechatauto-replica`, or the prebuilt Windows x64 Python 3.12 runtime from the repository's Release (it already includes `wechatauto-replica`, `uiautomation`, `numpy`, `Pillow`, `pywin32`, `comtypes`)
- A DSH host that provides `tools.guard` — the README notes this version of DSH has it, and without it only a prompt-level restriction remains
Installation steps
- 01
Install the plugin: `dsh plugin --profile desktop add github:thawline/dsh-wechat-bridge`, or use 设置 → 插件 → 安装 and enter `github:thawline/dsh-wechat-bridge`
$ dsh plugin --profile desktop add github:thawline/dsh-wechat-bridge
- 02
Restart DSH once, then open 设置 → 通到微信
- 03
Click 「装依赖」 (install dependencies) — either point the 「解释器」 field at an existing Python that already has `wechatauto-replica`, or let it build a virtualenv from system Python 3.10+
- 04
Without Python: unzip `wechat-bridge-runtime-win64-py312.zip` and put its `python` folder at `~/.dsh/wechat-bridge-runtime/python` so that `C:\Users\<you>\.dsh\wechat-bridge-runtime\python\python.exe` exists; the plugin finds it automatically
- 05
Click 「启动」 (start) and watch the log, then pick a 「默认会话」 and turn on 「总开关」
- 06
Leave 「试运行」 on at first and check that the 「该发什么」 log lines are right, then switch it off
Verify the integration
- The bridge log shows 「桥接已就绪」 after clicking 「启动」
- With dry-run on, the log records what would be sent; after turning dry-run off, a message sent to 文件传输助手 comes back as the agent's reply in WeChat
Rollback
- Turn off 「总开关」 — while it is off nothing is sent at all
- The bundled runtime is portable: the README says you can move the folder anywhere and deleting it is the uninstall
03DSH integration and capability boundaries
A DSH plugin plus a bundled Python bridge: the host half handles session binding, permissions and queues, the Python half reads WeChat's local SQLite database and sends via Windows UI Automation, and the two talk over local HTTP endpoints under `/dsh-wechat-bridge/`.
Inbound WeChat → DSH session
a WeChat private or group message, read from WeChat 4.x's local SQLite database (read-only, WeChat must be logged in)→the message turned into a turn in the bound DSH session, with group context lines and a per-chat or per-group binding
inbound images and stickers are written to disk first, then described by a vision model before entering the sessionvoice and video messages are still skippedOutbound DSH reply → WeChat, via UI Automation and the clipboard
the last reply of a completed agent turn, plus `[表情: description]` markers→text typed into the real WeChat window and sticker images pasted as picture messages; the `[表情: …]` marker itself is not sent as text
drives the actual WeChat window through UI Automation and the clipboardmay pull WeChat to the foreground on a failed send and retryimage sends are confirmed by checking WeChat's database for a new pictureTwo-layer permission model
an inbound allowlist of wxids / nicknames, and the 「主人」 (owner) list→an allowlist gate deciding who can start a turn at all, and a `tools.guard` gate deciding whether tools may run during that turn
non-owner turns have every tool call rejected before executionthe owner is always let through and does not need to be listed in the allowlistTwo tools and a local HTTP surface
natural language in chat, or HTTP calls to the `/dsh-wechat-bridge/*` endpoints→`wechat_block` for the do-not-reply list and `wechat_bind` for binding a chat to a session; receive / outbox / state / save / test-send / install / start / stop endpoints
config is written to `~/.dsh/wechat-bridge-config.json` and saved changes take effect within 30 seconds without a restarta virtualenv lives at `~/.dsh/wechat-bridge-venv`
04Who is it for? When not to use it?
Good for
- Windows users who already live in WeChat and want their local DeepSeek Harness agent reachable from the phone
- People who want friends or a group to try the agent but do not want their files and shell exposed
- Groups that want the agent present but rate-limited, with a searchable transcript afterwards
Not for
- Automating a personal WeChat account is not an official interface and carries a ban risk — the README advises starting with a secondary or test account.
- Windows only — UI Automation is a Windows-only API — and only one bridge may drive a given WeChat account at a time, otherwise windows and sends collide.
- A WeChat upgrade can break sending: `wechatauto` locates the control tree through an address table hard-coded per version, so after an upgrade check the log under the settings page first.
- By default anyone who can message that WeChat account can start a turn and spend your model quota — the tool guard only stops them touching the machine, so set the inbound allowlist if you want to close that layer.
05Compatibility, maintenance and safety notes
- Automating a personal WeChat account is not an official interface and carries a ban risk — the README advises starting with a secondary or test account.
- Windows only — UI Automation is a Windows-only API — and only one bridge may drive a given WeChat account at a time, otherwise windows and sends collide.
- A WeChat upgrade can break sending: `wechatauto` locates the control tree through an address table hard-coded per version, so after an upgrade check the log under the settings page first.
- By default anyone who can message that WeChat account can start a turn and spend your model quota — the tool guard only stops them touching the machine, so set the inbound allowlist if you want to close that layer.
MIT · README evidence only, latest release v1.3.4, no security audit; a modified fork of Yuuyuko-uu/dsh-wechat-bridge
06Frequently asked questions
How do I install the WeChat bridge into DeepSeek Harness?
Run `dsh plugin --profile desktop add github:thawline/dsh-wechat-bridge`, or open 设置 → 插件 → 安装 and enter `github:thawline/dsh-wechat-bridge`. Restart DSH once, then go to 设置 → 通到微信, click 「装依赖」 to install the Python side (point 「解释器」 at an existing Python with `wechatauto-replica`, or let it build a venv from Python 3.10+), click 「启动」 and wait for 「桥接已就绪」 in the log.
Do I need to install Python myself?
No — the repository's Release ships a ready-made Windows x64 Python 3.12 runtime with `wechatauto-replica`, `uiautomation`, `numpy`, `Pillow`, `pywin32` and `comtypes` already installed. Unzip `wechat-bridge-runtime-win64-py312.zip` and place its `python` folder at `~/.dsh/wechat-bridge-runtime/python`, and the plugin will find it without the 「解释器」 field or the 「装依赖」 button.
Can other people use my machine through WeChat?
Not through tools. When an external message arrives the plugin attaches a `tools.guard` to that session for the turn: if a non-owner woke it, every tool call is rejected before execution and only text replies pass, so 「帮我删掉 D 盘」 just gets refused. The catch is that by default anyone who can message the account can still start a turn and spend your quota — set the 入站白名单 to restrict who may trigger at all, and note the guard requires a host that provides `tools.guard`.
Will my group chats get spammed by the agent?
The README's defaults are conservative: at least 30 seconds between two replies in the same group, at most 6 replies per group per hour, and 6 lines of group context per wake-up. All three are configurable and 0 means unlimited. In groups the agent only joins when named (the aliases are hard-coded in `bridge/wechat_bridge.py`) or when the owner speaks.
What are the risks before I turn it on?
Personal WeChat automation is not an official interface and carries a ban risk, so the README suggests a secondary or test account first. It is Windows-only because UI Automation is Windows-only, and a WeChat upgrade can break sending because `wechatauto` finds the control tree via a per-version hard-coded address table. Do not run two bridges against the same WeChat account.
07Related DSH workflows
awesome-gpt-image-2
by freestylefly
Prompt-as-Code engine and template library for GPT-Image2, with 470+ reverse-engineered cases, 20+ production templates and distilled agent skills.
dsh-im
by xmanrui
把 IM 机器人接入 DSH：统一管理飞书/微信/钉钉/企业微信/QQ/Telegram/Discord/WhatsApp 八个渠道，扫码或凭据接入、流式回复与设置页管理
echobird
by edison7009
All-in-one AI desktop app: one-click install for Claude Code, Codex CLI, DeepSeek and more, manage AI accounts and quotas, smart model routing with auto-failover, voice input, and run local LLMs. Bonus: a Token Pet that grows as you use AI.
wegent
by wecode-ai
Plan, build, and deliver with an open-source, self-hostable AI workspace for coding, collaboration, and automation.
08Data and sources
宿主半边管会话绑定、权限、队列；Python 半边管读库、发送、日记。中间走本机 HTTP。
收消息：读微信 4.x 的本地 SQLite 数据库（只读，要求微信处于登录状态）
dsh plugin --profile desktop add github:thawline/dsh-wechat-bridge
This page is generated from the project’s public documentation, repository metadata and a structured parse of DSH Plugins; last verified on 2026-11-14. Found an error? Submit a correction.
Best DeepSeek Harness Plugins
Twelve plugins worth installing first — picked from the whole catalog, across every category.