Skip to main content
Examples use a wholly fictional cross-border marketplace merchant-recruitment scenario: onboarding information, product categories, target markets, missing documents and human follow-up. They are not drawn from real customers and state no platform policies, fees or admission promises.
The default entry is AI Workbench; see the workbench guide. The legacy Assistant Chat steps for #, / and images below apply only where that drawer and command list are actually available. Check the currently supported commands before sending. Chat does not automatically read the current page or selected window, or replace saving configuration.
MarsMind’s AI is a digital coworker that shares the work with you: it serves each conversation under the role you set, asks you when it is unsure, and your answers settle into knowledge it can use next time. This page walks you through that loop — how roles are set, where to give it commands, when it comes to ask, and how to respond.

When you’ll need this

  • You just connected a communication window and are about to let the AI serve customers: set who is who first, so it does not mistake one person for another or give the wrong answer.
  • You want to direct the AI yourself: start or pause its work, turn group chat replies on or off, organize records, or generate reports.
  • The AI forwarded a question it could not answer: understand why it is asking, how to respond, and what happens if you never reply.
  • Troubleshooting: when the AI answers a colleague as if they were a customer, or speaks in a group where it should stay quiet, go back to the role settings first.

The four roles the AI sees

How the AI handles a conversation depends on who the other side is in its eyes. These roles are maintained in Relationships; the four roles mean: Your own identity is bound by an authorization code: generate a Supervisor Authorization Code or a Colleague Authorization Code in AI identity access and send it to the AI account from your own phone. For the same AI account, the last code received wins — supervisor and colleague overwrite each other.
Relationships supports only logged-in WeChat and WeCom accounts; Website link, Mini Program, and Official Account windows do not have this contact and group list.

How to get there

This page spans several entry points that work together:
  • Chat with the AI and give commands: top bar, right side → “Assistant Chat”.
  • Set roles for contacts and groups: left menu → My Assistant → Relationships.
  • Bind your own identity: bottom-left “Account settings” → “AI identity access”.

Steps

1

Check who is who in Relationships

Entry: left menu → My Assistant → Relationships. First select the window you want to check under “Selected window” in the sidebar.In the contact and group list, check the “Relationship” column row by row. To correct a row, click “Change role” at the end of it and pick one of “Customer”, “Colleague”, “Muted”, or “Supervisor” in the “Change contact relationship” dialog, then click “Save relationship”. To change several at once, select them and click “Set as Customer”, “Set as Colleague”, “Mute”, or “Set as Supervisor” on the action bar.How to tell it worked: the “Relationship” column shows the result you wanted — pay special attention to “Supervisor”, which is the recipient for questions the AI forwards and reports it sends proactively.Note: batch buttons take effect immediately, with no confirmation step. If a newly added friend or group is missing from the list, click “Sync contacts” at the top first (one sync per day at most). If no window is available yet, the page shows “No communication window is available”; click “Go to channel login” to connect one first.
2

Make the AI recognize you with an authorization code (Supervisor / Colleague)

Bottom-left “Account settings” → “AI identity access”. Copy the Supervisor Authorization Code or the Colleague Authorization Code: the Supervisor code copies in the send-ready format with the # prefix (for example, #A1b2C3), while the Colleague code copies as the 6-character code itself. Open the chat with this AI account on your phone and send the copied content as-is.How to tell it worked: “Supervisor Authorization Code copied” appears (the colleague code shows “Colleague Authorization Code copied”); after you send it, the AI recognizes that phone account as a supervisor or a colleague.Note: for the same AI account, the last code received wins, and supervisor and colleague overwrite each other — to switch back to Supervisor, just send the Supervisor Authorization Code again. “Generate and save a new code” replaces the old code immediately, with no confirmation step; the old code stops working, so you must send the new one.
3

Chat and give commands in Assistant Chat

Click “Assistant Chat” on the right of the top bar to open the “Admin assistant chat” drawer, then type your message in the input box and send it to start chatting.
  • Type ”/”: natural-language suggestions appear, such as “Please help me organize today’s conversation records” and “Generate this week’s work report” — pick one or write your own.
  • Type ”#”: forced commands appear, such as #work (start work), #stop (pause work), and #help (view all commands) — if you cannot remember the commands, send #help first.
How to tell it worked: your message and the AI’s reply appear in the conversation; if sending fails, an error pops up (when there is no specific error message, it reads “Failed to send message, please try again”).
Fictional merchant-recruitment demo: The open "Admin assistant chat" drawer: the scope note at the top, with the message input box, image upload, and send button below
Note: the drawer header reads “This entry is not bound to the communication window selected on the page. It cannot confirm that window’s configuration is in effect.” — you can give commands and ask questions here, but do not use it to test how a specific communication window replies.
4

Answer when the AI comes to you for help

The AI proactively comes to you in three situations:
  • It cannot answer: when the AI cannot answer a customer’s question, it forwards a question summary to the supervisor — the “Suggestions” metric under Performance counts these forwards.
  • Key events: when a situation you wrote under Reply Rules → “Key Event Active Reporting” occurs, it reports proactively.
  • Guardrail hit: when a rule in Safety Guardrails has “Leader notification” set to “Notify Leader” and the rule is triggered, it sends a notification.
Reply through the authorized channel or hand unconfirmed merchant eligibility and document questions to a person. Submit reusable answers through knowledge generation and review after human confirmation.How to verify: the real channel question has an answer or a named follow-up owner, and newly submitted production knowledge is searchable in Knowledge Management.Note: to receive questions and reports, you must first be a “Supervisor” in the AI’s eyes — you have sent the Supervisor Authorization Code, or you are set as “Supervisor” in Relationships. The switch and content for “Key Event Active Reporting” must also be enabled and saved in Reply Rules.
5

Review the collaboration in reports and decide what to teach next

Left menu → Data Dashboard → Performance: watch how “Suggestions” (the number of question summaries the AI forwarded to supervisors) and “Learned Items” (the knowledge it learned automatically from conversations) change, and judge whether it is resolving things on its own or asking frequently — questions it asks about often are what to add to the knowledge base next.To understand why a reply was given the way it was, go to left menu → Data Dashboard → Detail Review and open an inquiry to check its “Reference Knowledge” and “Execution Steps”.How to tell it worked: “Suggestions” and “Learned Items” show data; in the records you spot-check, replies map to reference knowledge or execution steps.Note: Messages Received, Inquiries, and Replies overlap and cannot be added together; for the definition of any other metric, click “Metric Guide” at any time.

Checkpoints

  • The “Relationship” column in Relationships matches your intent, and the “Supervisor” recipients are clear.
  • The Supervisor or Colleague Authorization Code has been sent to the corresponding AI account, and it is the one currently in effect (after generating a new code, send it again).
  • Assistant Chat can send and receive messages; typing ”#” and ”/” shows the command and suggestion lists.
  • Merchant questions needing human follow-up have an owner; reusable answers were submitted and verified in stored knowledge, independently of test-list counts.
  • Suggestions and Learned Items show data in Performance.

Important notes

Authorization codes are identity credentials: send them only to the person being authorized and the corresponding AI account — never forward them to unrelated groups or people. “Generate and save a new code” replaces the old code immediately, with no confirmation step, and the old code stops working.
Authorization codes tell the AI “who you are”; Relationships show “who your contacts and groups are” — together they decide how the AI treats each conversation partner.
Assistant Chat is not bound to the communication window selected on the page, so it cannot confirm that a window’s configuration is in effect. To verify a configuration, send messages through a real channel, or drill into anonymized recruitment conversations in Detail Review.
Typing ”/” and describing what you need is faster than memorizing commands; if you cannot remember them, send #help to get the full list of commands the AI currently supports.
Needs answer (Test) uses local demo questions. Confirming, ignoring or changing categories does not write to the production knowledge base; refreshing restores the examples. Submit reusable answers through Knowledge Generation → Enter Existing Knowledge, then review and verify storage.

Frequently asked questions

What is “Suggestions”, and when will the AI ask me for help?

“Suggestions” counts the question summaries the AI forwards to supervisors. When the AI cannot answer a customer’s question, it forwards a summary to the supervisor; it also reports proactively on the key events configured under Reply Rules → “Key Event Active Reporting”; and when a rule in Safety Guardrails is set to “Notify Leader”, it notifies you as well.

The AI forwarded a question to me and I did not reply in time — what happens?

Needs answer (Test) uses local demo questions. Confirming, ignoring or changing categories does not write to the production knowledge base; refreshing restores the examples. Submit reusable answers through Knowledge Generation → Enter Existing Knowledge, then review and verify storage.

Why has the supervisor never received a proactive report from the AI?

Check three places in order: (1) whether the recipient is a “Supervisor” — either they have sent the Supervisor Authorization Code, or they are set as “Supervisor” in Relationships; (2) whether “Key Event Active Reporting” in Reply Rules is enabled and its content saved (click “Save Configuration”); (3) for a guardrail notification, whether the rule’s “Leader notification” is set to “Notify Leader”.

Can I use Assistant Chat to verify a communication window’s configuration?

No. It is not bound to the window selected on the page. To verify, send messages through a real channel, or open Detail Review and drill into a anonymized recruitment conversation to check its “Reference Knowledge” and “Execution Steps”.

A colleague keeps being answered as a customer — how do I fix that?

Use either fix: change their “Relationship” to “Colleague” in Relationships, or have the colleague send the Colleague Authorization Code to this AI account. Afterward, confirm the AI’s treatment has changed with a real message. Batch changes take effect immediately, and there is no undo.