Skip to content
DuDuClaw OSDuDuClaw OSDuDuClaw OS Docs

Thinking display in messaging apps

People who talk to an AI employee in Telegram, LINE, Slack and the other messaging apps can see that it is thinking, and colleagues who are bound to a dashboard account can also read a short, masked summary of the thinking. Customers see the timer only, group chats show the timer only by default, and website visitors see nothing.

The summary is the same text the dashboard chat shows (see Thinking display and Where thinking comes from): masked first (redaction tokens become 〔已遮蔽·category〕 and are never restored, credential-shaped words are replaced, the CONTRACT and guardrail checks run) and capped. It is never stored.

What each app shows

Section titled “What each app shows”
App While thinking When the answer is ready
Telegram, Discord, Slack, Microsoft Teams One status message “💭 思考中… 12 秒”, updated at most every 3 seconds, with the latest summary line for senders who may read it The same message is edited into “💭 思考了 12 秒” with the summary: an expandable quote in Telegram, a spoiler in Discord (mentions disabled), a small context block in Slack, the first 300 characters in Teams. The answer follows below it
LINE (1:1 chats) The LINE loading animation (renewed every 50 seconds) One small extra message “💭 思考了 12 秒” (+ the first 300 characters of the summary) placed before the answer in the same reply call. Nothing is added when the answer already uses all five messages of a reply, or when the answer goes out by push after the reply token expired: thinking never costs an extra push
LINE (groups) Nothing (LINE has no loading animation for groups) The same extra message, timer only by default
WhatsApp, Feishu, WeCom, DingTalk, Google Chat The existing “already received” status line One line “💭 思考了 12 秒” (+ the first 300 characters of the summary) at the end of the answer, above the “Powered by DuDuClaw” footer

“12 秒” is the time from the moment the message arrived until the answer was ready, the same number the dashboard chat shows. Nothing is shown when the model did not report any thinking. On the editable apps the “already received” status line is not posted while the thinking status shows; if the reply ends up empty or the turn is cancelled, the status message is removed. When an app refuses the status message three times in a row (or accepts it without returning its id), no more status messages are sent that turn. Teams gets the status as plain text, and the one-line footer on the other apps has mention and link syntax made inert, so a summary can never ping anyone or render a link. Labels follow [general] default_language (Traditional Chinese, English or Japanese).

Who sees what

Section titled “Who sees what”

config.toml:

[thinking]
channel_audience = "internal"     # internal | all | none
channel_group_display = "status"  # summary | status | off
Setting One-to-one chat Group chat
internal (default), sender is an Admin or bound to the answering employee timer + summary timer (channel_group_display)
internal, any other sender timer timer
all timer + summary timer (channel_group_display)
none nothing nothing
  • “Bound” means the sender’s channel identity is verified, belongs to an active dashboard account, and that account is an Admin or is bound to the employee that answers (the same binding the dashboard uses for that employee). An account bound only to other employees sees the timer. The binding is read from the user database on every message, so revoking it takes effect at once. A failed lookup counts as not bound.
  • channel_group_display is the most a group chat shows, whoever speaks. Set it to summary to let bound colleagues read summaries in groups too.
  • The stricter of [thinking] display and the answering employee’s [model] show_thinking caps everything: status means the timer only, off means nothing. An employee called with @ uses its own setting.
  • A config.toml that cannot be read or parsed, or an unknown value, shows nothing.
  • Website visitors (Odoo website live chat, the WebChat widget) never get thinking, whatever the settings.

Only senders who may read the summary make the runtime ask for thinking text (see Where thinking comes from); for everyone else requests stay exactly as before. Showing thinking does not change what a turn costs.

Not verified on real accounts: how Telegram renders the expandable quote and Discord the spoiler on every client, Slack’s context block in threads, the LINE extra message on a real phone, and Teams edits. The messages were checked against a local stand-in for each API.