dsh-weixin

> Part of the DSH ecosystem — channel plugins built on the DeepSeek Harness (DSH) plugin architecture.

Personal WeChat channel plugin for the DeepSeek Harness (DSH) — turn dsh into a WeChat bot with one long-lived Agent per chat peer (true rolling memory, flushed to disk).

How it works

This plugin talks directly to WeChat's official iLink Bot API (the same protocol behind Tencent's @tencent-weixin/openclaw-weixin gateway plugin). It uses that Tencent-maintained package as a plain protocol library — no OpenClaw gateway, no forwarding, no cloud relay — and replaces the gateway glue with DSH Agents:

your WeChat ⇄ Tencent iLink API ⇄ this plugin (getUpdates long-poll)
                                        └→ one resident DSH Agent per peer
                                           replies via sendMessage

• QR login on first boot: the QR link is printed to the log and written to <rootDir>/login-qr.txt (render it or open it in WeChat). The flow auto-refreshes expired QRs and retries.
• Credentials are stored under <rootDir>/state/openclaw-weixin/accounts/<bot-id>.json (0600), isolated from any OpenClaw installation via OPENCLAW_STATE_DIR.
• Inbound: every text message wakes the peer's Agent (created lazily, memory spans the process lifetime; sessions persist to the DSH session store).
• Model: whatever the host's default model selection is (agent-default-model in ~/.dsh/settings.yaml).
• Resilient: long-poll timeouts and transport errors back off exponentially (5s→120s) and reconnect.

Install

# 1. Put the bundle where DSH resolves bundles, e.g. the runtime install:
cp -R dsh-weixin ~/.dsh/runtime/node_modules/dsh-weixin
# (plus the protocol dependency, next step)

# 2. Install the iLink protocol library into the same runtime:
cd ~/.dsh/runtime
npm install @tencent-weixin/openclaw-weixin@2.4.6 openclaw
# (`openclaw` satisfies one utility import in the library's logger;
#  no gateway is started or used)

# 3. Create ~/.dsh/profiles/weixin/package.json:
# {
#   "name": "dsh-profile-weixin",
#   "private": true,
#   "dependencies": {},
#   "dsh": { "profile": { "bundles": ["@deepseek-ai/dsh-base", "dsh-weixin"] } }
# }
# plus empty cordis.yml (`[]`), cordis.patch.yml (`[]`), pnpm-workspace.yaml.

# 4. Boot:
dsh --profile weixin

On first boot it prints a QR link — scan it with your phone WeChat, confirm, and the bridge enters the poll loop. Subsequent boots reuse the saved token.

Configure (optional, via the profile's cordis.patch.yml)

- id: weixin-bridge
  config:
    rootDir: ~/.dsh/weixin-channel   # state + workspace root
    accountId: default               # logical account label
    loginTimeoutSeconds: 240         # per-attempt QR wait
    minBackoffSeconds: 5             # transport backoff bounds
    maxBackoffSeconds: 120
    maxReplyChars: 3800              # outgoing reply cap

Notes & limits

• One iLink connection per WeChat account globally — don't run another

WeChat gateway (OpenClaw, Hermes, hermesclaw…) on the same account while this bridge runs, or tokens will conflict (403s / dropped messages).

• Text messages only in v0.1 (the protocol lib also supports images/voice/

files — media handling is a natural next step).

• Self-sent messages don't reach the bot channel; test from another account.
• Reply latency = model turn time (tens of seconds typical); the iLink side

has no typing indicator wired up in v0.1.

License: MIT-style, do whatever you like; no warranty.