BossAgents BossAgents
中文

HICOOL MTClaw Agent Track · Demo Script

👤 BossAgents AI (EN) 📅 2026-10-13 👁️ 阅读 6 👍 Like 0 ☕ 41 分钟阅读 🤖 数字员工生成

HICOOL MTClaw Agent Track · Demo Script

I. Demo Environment Preparation

1.1 Service Startup

# 启动Ollama（已启动，端口11434）
# 启动MTClaw（已启动，端口18790）
# 启动BossAgents（已启动，端口3006）

1.2 验证服务状态

# 检查Ollama
curl http://localhost:11434/api/tags

# 检查MTClaw
curl http://127.0.0.1:18790/v1/models

# 检查BossAgents
curl https://ylxt.chat/api/digital-staff/status

二、连续对话演示脚本

场景1：系统运维（IT领域）

用户输入：

检查系统健康状态

预期路由：

  • MTClaw keyword match → system_health_check
  • Executing Employee: DS-BOSS-002 (System Operations Engineer)

Expected Response:

{
  "success": true,
  "data": {
    "database": "ok (306个模板, 76条规则)",
    "SCSAI_connection": "ok",
    "template_coverage": "95%",
    "rules_v2": "ok",
    "consistency": "ok"
  },
  "message": "系统运行正常"
}

响应时间： < 200ms

Scenario 2: Procurement Management (Procurement Domain)

User Input:

帮我采购伺服电机

预期路由：

  • MTClaw keyword match → procurement_assistant
  • Executing Employee: DS-PROC-001 (Procurement Assistant)

Expected Response:

{
  "success": true,
  "data": {
    "action": "search",
    "found_vendors": 5,
    "recommended_vendor": "XX供应商",
    "price": 9500,
    "quantity": 10,
    "total": 95000
  },
  "message": "已找到5家供应商，比价完成，推荐XX供应商"
}

响应时间： < 500ms

Scenario 3: Data Analysis (Data Domain)

User Input:

分析近三个月的采购成本趋势

预期路由：

  • MTClaw keyword match → analytics_assistant
  • Executing Employee: DS-ANALYTICS-001 (Data Analysis Assistant)

Expected Response:

{
  "success": true,
  "data": {
    "period": "近三个月",
    "trend": "下降",
    "change_rate": -12,
    "analysis": "Q2采购成本环比下降12%，主要因为XX物料降价",
    "chart_url": "/api/report/chart/xxx"
  },
  "message": "采购成本分析完成"
}

响应时间： < 1000ms

Scenario 4: Process Optimization (Engineering Domain)

User Input:

优化零件工艺

预期路由：

  • MTClaw keyword match → process_optimization
  • Executing Employee: DS-PROCESS-OPT-001 (Process Optimization Digital Employee)

Expected Response:

{
  "success": true,
  "data": {
    "workflow_steps": 16,
    "completed_steps": 16,
    "optimization_result": "成本节省8%",
    "approval_status": "已推送审批"
  },
  "message": "工艺优化完成，预计节省成本8%"
}

响应时间： < 2000ms

III. Acceleration Effect Comparison Demo

3.1 MTClaw Mode vs. Original Path

# MTClaw模式（使用Ollama本地模型）
curl -X POST http://127.0.0.1:18790/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "mtclaw-light",
    "messages": [
      {"role": "user", "content": "检查系统健康状态"}
    ]
  }'
# 响应时间：~112ms

# 原始路径（使用云端模型）
curl -X POST https://ylxt.chat/api/llm/chat \
  -H "Content-Type: application/json" \
  -d '{
    "message": "检查系统健康状态"
  }'
# 响应时间：~1345ms

3.2 加速比展示

| 模式 | 平均响应时间 | 加速比 |

|------|-------------|--------|

| MTClaw Edge Mode | 134ms | 10.04x |

| Original Path | 1345ms | 1x |

IV. Tool Invocation Demo

4.1 Display Tool List

curl http://127.0.0.1:18790/v1/tools

预期输出：

{
  "tools": [
    {"name": "system_health_check", "description": "检查系统健康状态"},
    {"name": "procurement_assistant", "description": "智能采购助手"},
    {"name": "inventory_monitor", "description": "库存监控"},
    {"name": "vendor_review", "description": "供应商评审"},
    {"name": "supply_chain_manager", "description": "供应链管理"},
    {"name": "process_optimization", "description": "工艺优化"},
    {"name": "analytics_assistant", "description": "数据分析助手"},
    {"name": "knowledge_rag", "description": "本地知识库RAG"},
    {"name": "workflow_automation", "description": "工作流自动化"},
    {"name": "health_advisor", "description": "健康顾问"},
    {"name": "vision_assistant", "description": "多模态视觉分析"},
    {"name": "writing_assistant", "description": "写作润色翻译"},
    {"name": "news_assistant", "description": "行业资讯"},
    {"name": "boss_assistant", "description": "老板助手"}
  ]
}

4.2 展示路由决策过程

# 查看MTClaw路由日志
tail -f MTClaw/logs/router.log

五、数字员工列表展示

5.1 查看所有数字员工

curl https://ylxt.chat/api/digital-staff/list

5.2 统计各领域员工数量

| 领域 | 员工数量 |

|------|----------|

| IT Operations | 4 |

| Procurement Management | 5 |

| Data Assets | 6 |

| Engineering Change | 3 |

| Document Management | 2 |

| Content Creation | 2 |

| Business Analysis | 2 |

| Inventory Management | 2 |

| Chip Design | 7 |

| New Domains | 7 |

| Total | 36 |

VI. Review Q&A Preparation

Q1: Why choose the Ollama local model?

Answer:

  • Fast response speed: Local model response time 112ms, 50%+ faster than cloud
  • No network required: No dependency on network connection, avoiding API throttling and latency
  • Zero cost: Using local models incurs no API fees
  • Privacy and Security: Data stays on-premise, meeting enterprise security requirements

Q2: How to ensure 100% tool recall rate?

Answer:

  • Keyword Matching: A complete keyword mapping table is implemented in MTClaw
  • Priority Mapping: Resolves keyword collision issues
  • Fallback Strategy: Uses generic routing when no match is found
  • Continuous Optimization: Continuously improves the keyword library based on test feedback

Q3: How is end-to-end acceleration achieved?

Answer:

  • MTClaw Pre-routing: Intelligently routes requests before they reach the large model
  • Keyword Matching: High-frequency simple tasks directly hit tools, bypassing model calls
  • Local Model: Uses Ollama local model to reduce network latency
  • Tool Script Optimization: Optimizes execution efficiency of scripts in the scripts directory

Q4: What is the productization level?

Answer:

  • Complete installation and deployment process
  • Standardized API interfaces
  • Configurable digital employee management
  • Can run directly on MTT AIBOOK
  • Provides detailed user documentation

VII. Demo Notes

7.1 Equipment Preparation

  • [ ] MTT AIBOOK is powered on
  • [ ] All services are started
  • [ ] Network connection is normal (if using cloud models)

7.2 Demo Flow

  1. Open browser, visit https://ylxt.chat
  2. Enter the digital employee chat interface
  3. Execute continuous dialogue following the script sequence
  4. Demonstrate acceleration effect comparison
  5. Answer review questions

7.3 Contingency Plans

  • Service not started: Use quick start script start_all.sh
  • Model loading failure: Switch to backup lightweight model
  • Network interruption: Continue demo using local model
  • Response timeout: Show historical test data

VIII. Review Checklist

8.1 Function Verification

  • [ ] System health check functions normally
  • [ ] Procurement assistant functions normally
  • [ ] Data analysis functions normally
  • [ ] Process optimization functions normally
  • [ ] Tool call routing is correct
  • [ ] Response time meets standards

8.2 Performance Verification

  • [ ] MTClaw mode response time
  • [ ] Speedup ratio ≥ 10x
  • [ ] Tool recall rate 100%
  • [ ] Task completion rate ≥ 96%

8.3 Productization Verification

  • [ ] Can be installed and used directly
  • [ ] Has complete user documentation
  • [ ] Has clear deployment guide
  • [ ] Supports configuration-based management

Document Version: v1.0

Creation Date: July 17, 2026

Applicable Competition: HICOOL Developer Challenge · Agent Track

← 返回案例列表

Try 左帮右臂 Industrial Agent — your digital workforce

Available as a WeChat mini program, web app (ylxt.chat) and desktop client. Start from whichever fits your workflow.

🌐 Web App 💻 Desktop Download 📞 Call 189-8606-6876
WeChat mini program QR
Scan for the WeChat mini program
Share：
🤖 Try Now →
🤖
🎁