Skip to main content

​
1. How AI Interacts with You When Uncertain

In daily customer communication, AI will try to complete replies independently. However, if it encounters uncertain situations, it won’t answer rashly but will consult you (Leader) immediately. It will send you two synchronized messages:
  1. Question Forward: The customer’s question forwarded as-is, ensuring you see the complete customer need.
  2. Thinking Draft: A reply draft generated by AI based on existing knowledge and context, for you to judge if it’s appropriate.
Your handling options:
  • Quote “Question Forward” and answer directly → AI will forward your reply to the customer as-is, ensuring precise and consistent expression.
  • Quote “Thinking Draft” and reply “1” or any content → Considered as approval, AI will send the draft to the customer.
  • Manually switch back to AI account to reply → Suitable for complex, sensitive, or scenarios requiring more personalized expression.
This mechanism both avoids incorrect answers affecting experience and allows AI to continuously learn your decision logic in practice.

​
2. How to Quickly Stop and Restart

In different scenarios, you may need to temporarily stop AI or have it re-enter working state.
  • Immediate pause: #stop → AI remains silent to customers but still responds to Leader messages.
  • Continue working: #work → AI resumes external service, returning to work mode.
After login, AI defaults to silent mode; it will only actively provide external service after you input #work. This design prevents unauthorized interruptions. Additionally, you can set silent whitelist/blacklist for specific groups or contacts in the admin backend for fine-grained control over which objects AI can respond to.

​
3. How to Teach It More and Keep It Getting Smarter

AI’s core value lies in continuous learning and gradually adapting to your business. Its main learning sources include:
  • Colleague and customer communication → Learning real business context and expression habits.
  • Leader’s direct guidance → Through your confirmation, denial, and rewriting, forming more robust judgment standards.
  • Backend knowledge base management → Systematic maintenance and updates, ensuring knowledge is traceable and governable.
  1. Have colleagues complete authentication (#auth → #peer) to avoid being misjudged as customers, ensuring AI correctly identifies internal knowledge and learns accordingly.
  2. Add AI to more real business groups to absorb typical cases and accelerate coverage of common problem domains.
  3. Engage in more “quote-style” guidance with AI; your rewrites and comments will be deposited as reusable learning materials.

​
Precise Knowledge Addition

Besides passive automatic learning, we recommend actively “feeding knowledge” to establish a more accurate and stable knowledge framework. Method 1: Backend Batch Management
  • Perform structured upload, hierarchical classification, and version management at admin.marsmind.cc / .co, suitable for large-scale knowledge governance.
Method 2: Single Entry (#chatknowledge)
  • Suitable for immediate, precise supplementation of hot knowledge points.
  • Operation steps:
    1. Use Leader account to input #chatknowledge (or #培训模式)
    2. Input customer question
    3. Input standard answer
    4. Takes effect after AI confirms recording
Please try to use real customer tone and supplement necessary context (object, limiting conditions, examples). This will significantly improve matching accuracy and transferability.

​
4. How It Responds in Groups (Group Chat Intelligent Response)

When AI is invited to join group chats, it will follow the “moderate participation” principle:
  • If customer questions appear and no one responds for about 3 minutes, AI will attempt to supplement answers;
  • If uncertain whether it can answer accurately, AI will first consult the Leader to avoid misleading.
Function switches:
  • Enable: #enable_hot_assistant
  • Disable: #disable_hot_assistant
When enabled, AI will continuously track and analyze group information, which may increase package usage. Please control based on scenarios and budget. The admin backend supports batch setting member identities (Leader/Peer/Customer) to avoid one-by-one command confirmation and reduce group misfire probability.

​
5. Using Daily/Weekly Reports

AI can generate periodic work reports to help teams synchronize information and review.
  • Daily report on/off: #enable_daily_report / #disable_daily_report
  • Weekly report on/off: #enable_weekly_report / #disable_weekly_report
Daily reports are suitable for high-frequency monitoring and troubleshooting; weekly reports are suitable for periodic review and trend tracking. You can flexibly enable based on management preferences.

​
6. Common Command Quick Reference

CommandFunctionTypical Scenario
#helpView all commandsWhen forgetting syntax/switch names
#auth {verification code}Identity verificationBefore executing role-related commands
#leaderSet as LeaderRecommend you serve as AI Leader
#peerSet as colleagueAvoid being judged as customer
#management {verification code}Set as managementRequires management password
#workStart workingActive external service
#stopPause workDebugging/temporary silence
#enable_hot_assistantEnable group responseSupplement answers when no one responds in group
#disable_hot_assistantDisable group responseControl usage/avoid misfires
#enable_daily_report / #disable_daily_reportDaily report on/offFrequent synchronization
#enable_weekly_report / #disable_weekly_reportWeekly report on/offPeriodic review
#chatknowledgeSingle knowledge entryImmediately supplement key knowledge points

​
7. Collaboration Best Practices and Support

Best Practices (Direct Reference)
  1. First familiarize yourself with basic commands and common operations, let yourself and your team gradually get used to interacting with AI, understanding when it needs your confirmation and when it can complete tasks independently.
  2. After becoming familiar, configure necessary identity relationships and permission boundaries based on actual scenarios. Clarify who is Leader, who is Peer, and pre-set groups or personnel that need to be silenced in the backend to avoid AI mistakenly sending to inappropriate targets.
  3. Recommend starting with a small number of core groups or customers for pilot operation, continuing for 1-2 weeks. Observe knowledge coverage, response effectiveness, and colleague onboarding, adjust based on results, then gradually expand to more groups and business scenarios.
  4. Long-term goal is to let people and AI gradually form stable collaborative relationships. You and your team provide commands, feedback, and scenarios, while AI continuously accumulates through execution and learning. Through regular backend optimization and knowledge base supplementation, make this collaboration increasingly smooth, truly becoming a reliable work partner.
Support and Contact We sincerely hope you can truly treat MarsMind AI as a colleague to collaborate with, like him, guide him, and gain continuous value from it. If you have any questions or suggestions during the collaboration process, please feel free to contact us: