SCSAI Digital Employee Platform — Rule Engine SOUL

Version: 3.0

Update Date: 2026-05-09

Core Concept: Natural Language → Keyword Recognition → Rule Matching → Tool Invocation → Result Return

Ultimate Goal: A Self-Operating, Self-Evolving Industrial Digital Employee System

I. System Soul (SOUL)

1.1 What is This

SCSAI is a rule-driven digital employee platform. It is not a simple chatbot, but a closed-loop business operating system:

``

Boss/Employee issues a command (natural language)

↓

Keyword Recognition Engine (L1: Pure rules, no LLM required)

↓

Rule Template Matching (ai-agent-rules.json)

↓

┌─ need_llm=false → Directly invoke API to execute

└─ need_llm=true → LLM-assisted understanding + Execute after confirmation

↓

Execution result returned (Frontend/Feishu/Automatic operation)

↓

System self-learning and self-optimization

`

1.2 Three-Level Intelligence Strategy

| Level | Trigger Condition | Processing Method | Latency | Cost |

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

| L1 Rule Direct | Keywords hit rules with need_llm=false | Directly call tool API |

| L2 Rule+LLM | Keywords hit rules with need_llm=true | LLM extracts parameters → User confirmation → Execution | 1-3s | Low |

| L3 Pure LLM | Keywords hit no rules | LLM understands intent → Reasons over tools → Confirmation → Execution | 3-10s | Medium |

Core principle: Never call LLM for tasks solvable by rules; never escalate to L2/L3 if L1 can handle.

II. Rule Template System (RULES)

2.1 Rule Template Structure

Each rule is a "soul fragment," defining how the system understands a category of instructions:

`json

{

"id": "Unique rule identifier",

"name": "Rule name (Chinese)",

"keywords": ["Keyword1", "Keyword2"],

"need_llm": false,

"action": {

"tool": "Tool name",

"params": { "parameter name": "parameter value template" },

"required_fields": ["Required fields"],

"SCSAI_type": "SCSAI object class name (optional)",

"SCSAI_select": "SCSAI query fields (optional)"

},

"examples": ["Example command 1", "Example command 2"]

}

`

2.2 Rule Matching Algorithm

`

Input: User message "Create supplier Huawei, phone 12345678"

↓

1. Convert to lowercase: "create supplier huawei, phone 12345678"

↓

1. Iterate through all rules, calculate the keyword hit count for each rule:

• create_vendor: "create"✓ "supplier"✓ "vendor"✓ → score 3
• create_project: "create"✓ → score 1
• query_vendors: "supplier"✓ "vendor"✓ → score 2

↓

1. Highest-scoring rule: create_vendor (score 3)

↓

1. need_llm=true → Attempt to extract parameters from the message:

• name: "Huawei" (extracted from message)
• phone: "12345678" (extracted from message)

↓

1. Return confirmation: "I understand you want to create a supplier: Huawei, confirm execution?"

↓

1. User confirms → Call create_vendor tool → Execute

`

2.3 Complete Rule List (47 items)

#### ? Query Rules (L1 Direct Pass, No LLM Required)

| ID | Rule Name | Keywords | Tool | Description |

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

| query_item_types | Query Object Class List | object class/list/ItemType/type | search_item_types | View all object class definitions in the system |

| query_inventory | Query Inventory | inventory/query/material/stock | list_data(inventory_items) | View inventory material list |

| query_projects | Query Projects | project/list/progress/project | list_data(projects) | View project list and progress |

| query_quality | Query quality data | Quality/Inspection/NCR/Pass Rate | list_data(inspections) | View quality inspection records |

| query_vendors | Query vendors | Supplier/Vendor/Manufacturer/List | list_data(vendors) | View vendor list |

| query_documents | Query documents | Document/Drawing/Document/File | list_data(documents) | View document list |

| query_cad | Query CAD drawings | CAD/Drawing/cad | list_data(documents) | View CAD drawings |

| query_customers | Query customers | Customer/Customer/Customer List | list_data(customers) | View customer list |

| query_work_orders | Query work orders | Work Order/Production/Work_Order | list_data(orders) | View production work orders |

| dashboard | Data overview | Overview/Statistics/Dashboard/Data | dashboard_summary | Dashboard data summary |

| compare_item_types | Comparison Object Types | compare/comparison/difference/diff | compare_item_types | Local vs. SCSAI Difference Comparison |

| system_health | System Health Check | health/status/system/check | system_health | Database/SCSAI/LLM Status |

| export_data | Export Data | export/download/export/excel | export_data | Export Module Data |

#### ? System Operation and Maintenance Rules (L1 Direct)

| ID | Rule Name | Keywords | Tool | Description |

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

| optimize_item_types | Batch Optimize Object Types | optimize/object type/attribute/completion | optimize_item_types | Automatically Complete Missing Attributes |

| import_item_types | Import Object Classes | Import/Sync/Object Classes/AML | import_item_types | Import from AML file |

| repair_item_type | Repair Object Classes | Repair/Object Classes/Integrity | repair_item_type | Repair object class definitions |

#### ✏️ Create Class Rules (L2 requires LLM assistance for understanding)

| ID | Rule Name | Keywords | Tool | SCSAI Type | Required Fields |

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

| create_vendor | Create Vendor | Create/Add/Vendor/Manufacturer | create_vendor | Vendor | name |

| create_project | Create Project | Create/Add/Project | create_project | Project | name |

| create_ecr | Create Change Request | Create/Add/Change Request/ECR/Change | create_SCSAI_item | ECR | name |

| create_ncr | Create Nonconformance Report | Create/Add/Nonconformance/NCR/Quality | create_ncr | NCR | title |

| create_purchase_order | Create Purchase Order | Create/Purchase/PO | create_SCSAI_item | pop_purchase_order | name |

| create_sales_order | Create Sales Order | Create/Sales/SO/Order | create_SCSAI_item | sop_sale_order | name |

| create_work_order | Create Work Order | Create/Work Order/Production Order | create_SCSAI_item | Work_Order | name |

| create_document | Create Document | Create/Document/Upload | create_SCSAI_item | Document | name |

#### ? Inventory Operation Rules (L2 requires LLM assistance for understanding)

| ID | Rule Name | Keywords | Tool | SCSAI Type | Required Fields |

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

| inventory_outbound | Inventory Outbound | Outbound/Picking/Shipment/Deduct Inventory | inventory_transaction | stk_issue | part_number, quantity |

| inventory_inbound | Inventory Inbound | Inbound/Receipt/Inbound Order | inventory_transaction | stk_issue | part_number, quantity |

#### ?️ Tagging Rules (L2 requires LLM assistance for understanding)

| ID | Rule Name | Keywords | Tool | Required Fields |

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

| toggle_core | Set as Core Object | Core/Business Object/Tag | toggle_core_business | item_type_name |

#### ? Analysis Rules (L2 requires LLM assistance for understanding)

| ID | Rule Name | Keywords | Tool | Required Fields |

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

| analyze_trends | Trend Analysis | Trend/Analysis/Statistics/Report | analyze_trends | module |

III. Tool Registry (TOOLS)

3.1 Tool Definition

Each tool serves as the "hands and feet" of a rule template, defining the specific execution logic:

| Tool Name | Description | Required Parameters | Data Source |

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

| search_item_types | Search Object Classes | None | SQLite |

| get_item_type_detail | Get Object Class Details | name | SQLite |

| list_data | Query Business Data List | module | SQLite |

| create_vendor | Create Vendor | name | SQLite + SCSAI |

| create_project | Create Project | name | SQLite |

| create_ncr | Create NCR | title | SQLite |

| create_quality_inspection | Create Quality Inspection | title, inspection_type | SQLite |

| create_inventory_item | Create Inventory Item | part_number, part_name | SQLite |

| inventory_transaction | Inventory Transaction | part_number, transaction_type, quantity | SQLite |

| create_SCSAI_item | General SCSAI Object Creation | item_type, name | SQLite + SCSAI |

| optimize_item_types | Batch Optimize Object Classes | None | SQLite |

| repair_item_type | Repair Object Class | name | SQLite |

| toggle_core_business | Set Core Object Flag | item_type_name, is_core | SQLite |

| import_item_types | Import object classes from AML | source | AML file + SQLite |

| compare_item_types | Compare object class differences | None | SQLite + SCSAI |

| dashboard_summary | Dashboard summary | None | SQLite |

| system_health | System health check | None | SQLite + SCSAI + LLM |

| analyze_trends | Trend analysis | module | SQLite |

| export_data | Data export | module | SQLite |

3.2 Tool Execution Flow

`

executeTool(toolName, params)

↓

1. Find tool in TOOLS array

↓

1. Validate required parameters (required)

↓

1. Call tool.handler(params)

↓

1. Return { type: 'result', success: true, data: ... }

or { type: 'error', message: '...' }

`

IV. SCSAI Object Class Mapping

4.1 PLM Core Object Classes

4.2 SCIOT Extended Object Classes

4.3 SCIOT Object Relationship Types

Real relationship types discovered from ItemType.xml (non-SCSAI standard names):

| Relationship Type | Query Method | Description |

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

| Property | | Direct relationship |

| Method | | Direct relationship |

| View | | Direct relationship |

| Allowed Permission | | Direct relationship |

| Server Event | | Direct relationship |

| Item Action | | Direct relationship |

| RelationshipType | | Queried via source_id |

V. Closed-Loop Execution Example

Example 1: L1 Direct Pass — "View Inventory"

`

User: "View Inventory"

↓

Keyword match: "Inventory"✓ "Query"✓ → query_inventory (score 2, need_llm=false)

↓

Direct execution: list_data({ module: "inventory_items" })

↓

SQL: SELECT * FROM inventory_items ORDER BY created_at DESC LIMIT 50

↓

Return: { type: 'result', data: { total: 156, items: [...] } }

↓

Frontend displays inventory list

`

Example 2: L2 Rule + LLM — "Create vendor Huawei, phone 12345678"

`

User: "Create vendor Huawei, phone 12345678"

↓

Keyword matching: "create"✓ "vendor"✓ "vendor"✓ → create_vendor (score 3, need_llm=true)

↓

Parameter extraction: name="Huawei", phone="12345678"

↓

Return confirmation: { type: 'confirm', message: "I understand you want to create vendor: Huawei\n\nConfirm execution?" }

↓

User clicks "Confirm"

↓

Execute: create_vendor({ name: "Huawei", phone: "12345678" })

↓

1. SQLite: INSERT INTO vendors (id, name, phone, ...) VALUES (...)
2. SCSAI: POST /server/odata/Part → Create Vendor object

↓

Return: { type: 'result', message: "Vendor created successfully (synced to SCSAI)" }

`

Example 3: Feishu Remote Command — "Inventory Alert"

`

Boss in Feishu group @Digital Assistant: "Inventory Alert"

↓

Feishu event push → POST /api/feishu/webhook

↓

handleMessageEvent → aiAgent.processMessage("Inventory Alert")

↓

Keyword matching: "inventory"✓ "alert"✗ → query_inventory (score 1, need_llm=false)

↓

Execute: list_data({ module: "inventory_items" })

↓

Feishu reply: "Current inventory alert items total 12:\n1. R001 Resistor - Current 0 Safety 100\n2. ..."

`

VI. AI Inspection System (Self-Evolution Core)

6.1 Inspection Architecture

The AI Inspection System is the core engine for platform self-evolution. It continuously detects data quality issues in SCSAI object classes, automatically fixes and validates them:

`

Inspection (1,197 object classes) → Problem identification (5 rule categories) → One-click fix (SSE real-time progress) → Automatic re-inspection verification

`

6.2 Inspection Rules (5 rules, all auto-fixable)

6.3 Technical Points for Fixing

• Serial execution + 200ms delay: Avoid SCSAI database deadlock
• Retry mechanism: Automatic deadlock retry, up to 2 times
• SSE real-time push: Push to frontend immediately after each fix item completes
• Data separation: SSE only pushes summary; full issues fetched via JSON API (to avoid frontend lag from large data)
• Authentication passthrough: Frontend passes current logged-in user's authentication to server, fixes executed using that user's permissions

6.4 API Interfaces

Seven. Digital Employee Team

7.1 Current On-Duty Employees (4 persons)

7.2 LLM Brain Capabilities

Eight. Self-Operation & Self-Evolution Roadmap

Phase One: Rule-Driven (✅ Completed)

• ✅ 47 rule templates covering core business
• ✅ 42 utility functions implementing business operations
• ✅ Three-level intelligent strategy (L1/L2/L3)
• ✅ Feishu bot remote commands
• ✅ Keyword matching + LLM fallback
• ✅ AI inspection system (5 rules, SSE real-time fixes)
• ✅ 4 digital employees working autonomously
• ✅ SCSAI SOAP API integration
• ✅ Dual-write data architecture (SCSAI + SQLite)

Phase 2: Self-Learning (? In Progress)

• ? Operation Log Analysis → Automatic Discovery of High-Frequency Commands → Automatic Generation of New Rules
• ? User Feedback Collection → Optimization of Rule Keyword Weights
• ? Rule Conflict Detection → Automatic Priority Adjustment
• ? AML Change Monitoring → Automatic Synchronization of Object Class Definitions
• ? Inspection Rule Expansion (Relationship Types/Lifecycle/Permissions/Naming Conventions)

Phase 3: Self-Operation and Maintenance (? Planned)

• ✅ Scheduled Health Check → Data Scribe Already Implemented
• ? Inventory Alert → Automatic Triggering of Procurement Suggestions
• ? Quality Trend Analysis → Automatic Initiation of Corrective Actions
• ? System Anomaly Detection → Automatic Diagnosis and Repair
• ? Object Class Integrity Automatic Inspection → Automatic Repair (AI Inspection System Basic Version Already Implemented)

Phase 4: Self-Evolution (Long-term)

• ? Business Process Mining → Automatically Discover Standard Processes
• ? Rule Auto-generation → Learn New Rules from Historical Operations
• ? Cross-Module Linkage → Automatically Trigger Related Operations
• ? Predictive Maintenance → Proactively Identify Potential Issues

IX. Core Business Module Mapping

9.1 Business Module → SCSAI Object Class → Rule/Tool Mapping

9.2 Business Modules to be Improved

10. Known Limitations and Improvement Directions

10.1 Current Limitations

1. Simulated Implementation: Approximately 13 advanced tools (visual inspection, equipment prediction, process optimization, etc.) still use simulated data.
2. Insufficient Inspection Rules: Only 5 basic rules, lacking checks for relationship types, lifecycle, permissions, etc.
3. Data Synchronization Limitations: Only 6 SCSAI object types are synchronized, missing BOM/Project/Customer, etc.
4. Insufficient Digital Employee Coverage: Only 4 digital employees cover 47 rules.
5. Lack of Testing Framework: No unit tests or integration tests.

10.2 Near-term Improvement Directions

1. Expand Inspection Rules: Add rules for relationship type integrity, attribute label localization, naming conventions, etc.
2. Add New Digital Employees: Quality Inspector, Inventory Manager, Production Scheduler.
3. Business Loop Closure: Inspection Discovery → Automatic Repair → Verification → Continuous Monitoring.
4. Large Model Integration: Combine inspection repair with LLM rule templates for intelligent repair decision-making.

11. How to Extend

7.1 Adding New Rules

Edit server/ai-agent-rules.json and add to the rules array:

`json

{

"id": "your_rule_id",

"name": "Rule Name",

"keywords": ["keyword1", "keyword2"],

"need_llm": false,

"action": {

"tool": "corresponding_tool_name",

"params": {}

},

"examples": ["Example Instruction"]

}

`

7.2 Adding New Tools

Edit server/ai-agent.js and add to the TOOLS array:

`javascript

{

name: 'your_tool_name',

description: 'Tool description',

parameters: {

type: 'object',

properties: {

param1: { type: 'string', description: 'Parameter description' }

},

required: ['param1']

},

handler: async (params) => {

// Implement your business logic here

return { result: 'success' };

}

}

`

7.3 Add a New SCSAI Object Class

1. Create an AML definition file for the ItemType in SCSAI/SCIOT
2. Add corresponding creation/query rules in ai-agent-rules.json
3. Add corresponding tools in ai-agent.js if special logic is required
4. The system will automatically synchronize the new object class via import_item_types

12. File Index