WeCom Integration

Talk to your Kiro Crew agent from WeCom — through a WeCom (企业微信) AI bot. Create the bot in your WeCom console, drop in two values, and you're chatting. Replies stream back live.

WeChat vs. WeCom. Kiro Crew connects through WeCom (企业微信), the work edition, using its AI-bot API. It does not sign in to a personal WeChat account — people message the bot from inside WeCom.

Like Telegram, the connection is outbound-only: Kiro Crew opens a secure WebSocket to WeCom, so there's no callback URL or open port to manage.

The easy way: just ask Kiro Crew

You don't have to edit anything by hand. In any Kiro Crew session — the dashboard, Slack, or the CLI — say something like "set up the WeCom channel." Kiro Crew tells you where to create the WeCom AI bot, then writes your Bot ID and Secret into ~/.kiro/crew/.env and config.json and restarts the gateway for you. You just paste the two values when it asks.

Prefer to wire it up yourself? The manual steps are below.

Quick start

You'll need a running gateway (kirocrew gateway) and admin access to your WeCom console.

1. Create an AI bot — in the WeCom admin console, open 应用管理 → AI 智能体 and create a bot. Its settings page shows a Bot ID and a Secret.
2. Note the userids — every WeCom member has a userid (账号). Collect the ones you want to let in.
3. Save the credentials to ~/.kiro/crew/.env:

   WECOM_BOT_ID=your-bot-id
   WECOM_SECRET=your-bot-secret
   

4. Turn it on in ~/.kiro/crew/config.json:

   "wecom": {
     "enabled": true,
     "allowed_users": [{ "userid": "zhangsan", "name": "Zhang San" }]
   }
   

5. Restart, then say hi:

   kirocrew restart
   

Message the bot in WeCom and it answers. If it stays quiet, look for WeCom WS connected and subscribed in the gateway log and confirm your userid is allowed.

Those two values — Bot ID and Secret — are all Kiro Crew needs. There's no corp ID, agent ID, callback URL, or AES key to wire up. Good to know about today's WeCom channel: it renders no tappable buttons, so a list of choices arrives as a numbered list you answer by typing. Tappable cards do exist in the AI-bot API; they are not wired up yet.

Access control

Kiro Crew runs on your machine, with your files and credentials. So it only talks to the owner and the userids you name.

• Authorized senders: the owner (KIROCREW_OWNER_ID) plus anyone listed in allowed_users. With no owner and an empty list, nobody gets in.
• Whole-company access: set "allow_all_users": true (or flip Allow all organization members in Settings → WeCom) to skip listing each userid. This is an explicit opt-in — an empty list never means "everyone" — and it works because a WeCom AI bot is only reachable inside your own org tenant. Messages without a userid are still dropped.
• Direct messages only — one conversation per userid. A message sent to the bot in a WeCom group is refused and recorded in the audit log, even from an allowed sender. Your session is keyed to your userid, so answering in a group would replay that private conversation's history and tool output to everyone in the room. Group support needs its own per-group sessions and group allow-list.
• Anyone else is quietly dropped and recorded in the audit log.

Sending files and screenshots

Send an image, a file, or a voice note and the agent sees it. Images and files are downloaded and decrypted from WeCom's CDN; a voice note uses WeCom's own transcript, so nothing extra has to be installed. A screenshot with a caption works too — the caption comes through with the picture, and a picture with no caption at all is still delivered rather than ignored. A caption that happens to look like a command is treated as text: attach a photo captioned /new and you get an answer about the photo, not a reset conversation and a lost picture. Send the command in its own message to run it.

Per-item ceilings follow WeCom's own: 10 MB per image, 20 MB per file (including an audio file you attach, which is transcribed locally), 10 items per message. A voice note has no size ceiling here because its bytes are never fetched — WeCom sends the transcript. Anything refused is reported rather than silently dropped. Sending a file to you is not wired up yet, so if the agent produces an image you'll get its path rather than the picture.

Commands

• /new (or 新对话, 清空) — start a fresh conversation
• /compact (or 压缩) — free up room when the context fills
• /stop (or /cancel, 停止) — stop the reply that's running
• /link / /unlink — mirror the dashboard's replies for this conversation here
• /help (or 帮助) — show the command list

While a reply is running, prefix a message to control it:

• /steer <message> — fold it into the running reply now
• /queue <message> — not supported here; you'll be asked to resend instead

Approving tools. WeCom has no approve/deny buttons, so a tool the agent wants to run is declined unless auto-approve is already on. Turn it on from the dashboard (or Slack) — it is one grant for the whole machine, it expires on its own, and tools your policy denies stay blocked either way. There is deliberately no WeCom command for it: one switch with one answer beats a per-channel copy.

Send commands in your direct chat with the bot. A leading @mention is tolerated and stripped before the command is matched, so @Kiro /new also works — but see "Who can reach it": messages sent in a WeCom group are refused, so commands only take effect in a direct chat.

Settings reference

Everything lives in the wecom section of config.json:

Setting	Default	What it does
enabled	false	Turns the channel on
allowed_users	[]	{ "userid", "name" } entries allowed to chat (empty = owner only)
allow_all_users	false	Allow every organization member to DM the bot
soft_threshold_pct	80	Context % where the bot suggests /compact
hard_threshold_pct	95	Context % hard cutoff
ws_url	wss://openws.work.weixin.qq.com	WeCom AI-bot endpoint
session_folder	""	Folder that WeCom sessions are filed under

Credentials go in ~/.kiro/crew/.env: WECOM_BOT_ID and WECOM_SECRET.

Delivery destinations

Dashboard delivery uses target_id values user:<userid>. Named users are shown as unavailable until they message the bot after the gateway starts; with allow_all_users, only those warmed direct-message peers are listed. Each destination is revalidated against the current authorization policy before delivery.

This channel is not an owner-DM target. Because the peer list is learned from inbound traffic, "the owner" cannot be resolved to a known person here — a send_message aimed at this channel would risk delivering to whoever last messaged the bot, so the channel is excluded from that routing entirely. Named dashboard destinations above still work; only the unqualified owner DM does not.

If something's off: no reply usually means the sender's userid isn't allowed or enabled is false; a missing channel started line means a credential is unset; if it connects and then goes quiet, the bot may have been removed from the WeCom console — re-add it and restart.

Related docs

• Channel capabilities: the ten-channel matrix — streaming, buttons, uploads, reply length, approval timeout
• Slack Integration
• Telegram Integration
• Getting Started