Roles & Permissions

Introduction

AI employee permission management consists of two levels:

1. AI Employee Access Permissions: Control which users can use which AI employees
2. Data Access Permissions: How AI employees apply permission controls when processing data

This document details the configuration methods and working principles of these two types of permissions.

Configuring AI Employee Access Permissions

Setting Available AI Employees for Roles

Go to the User & Permissions page, click the Roles & Permissions tab to enter the role configuration page.

Select a role, click the Permissions tab, and then click the AI employees tab. This will display the list of AI employees managed in the AI employees plugin.

Click the checkbox in the Available column of the AI employee list to control whether the current role can access that AI employee.

Data Access Permissions

When AI employees process data, the permission control method depends on the type of tool used:

System Built-in Data Query Tools (Follow User Permissions)

The following tools strictly follow the current user's data permissions for data access:

How It Works:

When AI employees call these tools, the system will:

1. Identify the current logged-in user's identity
2. Apply the data access rules configured for that user in Roles & Permissions
3. Return only the data that the user has permission to view

Example Scenario:

Suppose salesperson A can only view customer data they are responsible for. When they use AI employee Viz to analyze customers:

• Viz calls Data source query to query the customer table
• The system applies salesperson A's data permission filtering rules
• Viz can only see and analyze customer data that salesperson A has access to

This ensures that AI employees cannot bypass the user's own data access boundaries.

Workflow Custom Business Tools (Independent Permission Logic)

Business query tools customized through workflows have permission control independent of user permissions, determined by the workflow's business logic.

These tools are typically used for:

• Fixed business analysis processes
• Pre-configured aggregate queries
• Cross-permission boundary statistical analysis

Example 1: Overall Analytics (General Business Analysis)

In the CRM Demo, Overall Analytics is a template-based business analysis engine:

Workflow:

flowchart TD
    A[AI Employee Receives Task] --> B[Calls Overall Analytics Tool]
    B --> C[Passes collection_name Parameter]
    C --> D[Workflow Matches Corresponding Analysis Template]
    D --> E[Executes Pre-configured SQL Query]
    E --> F[Returns Business Analysis Data]
    F --> G[AI Employee Generates Charts and Insights]

Key Characteristics:

• Any user calling this tool will get the same business perspective
• Data scope is defined by business logic, not filtered by user permissions
• Suitable for providing standardized business analysis reports

Example 2: SQL Execution (Advanced Analysis Tool)

In the CRM Demo, SQL Execution is a more flexible but strictly controlled tool:

Security Recommendations:

1. Limit Scope: Only enable in management block tasks
2. Prompt Constraints: Clearly define query scope and table names in task prompts
3. Workflow Validation: Validate SQL statements in workflow to ensure only SELECT operations are executed
4. Audit Logs: Record all executed SQL statements for traceability

Example Configuration:

Task Prompt Constraints:
- Only query CRM-related tables (leads, opportunities, accounts, contacts)
- Only execute SELECT queries
- Time range limited to the last 1 year
- Return results limited to 1000 records

Permission Design Recommendations

Choose Permission Strategy by Business Scenario

Multi-layer Protection Strategy

For sensitive business scenarios, it is recommended to adopt multi-layer permission control:

1. AI Employee Access Layer: Control which roles can use which AI employees
2. Task Visibility Layer: Control task display through block configuration
3. Tool Authorization Layer: Verify user identity and permissions in workflows
4. Data Access Layer: Control data scope through user permissions or business logic

Example:

Scenario: Only finance department can use AI for financial analysis

- AI Employee Permissions: Only finance role can access "Finance Analyst" AI employee
- Task Configuration: Finance analysis tasks only display in finance modules
- Tool Design: Finance workflow tools verify user department
- Data Permissions: Finance table access permissions only granted to finance role

FAQ

Q: What data can AI employees access?

A: Depends on the tool type used:

• System built-in query tools: Can only access data that the current user has permission to view
• Workflow custom tools: Determined by workflow business logic, may not be limited by user permissions

Q: How to prevent AI employees from leaking sensitive data?

A: Adopt multi-layer protection:

1. Configure AI employee role access permissions to limit who can use them
2. For system built-in tools, rely on user data permissions for automatic filtering
3. For custom tools, implement business logic validation in workflows
4. Sensitive operations (such as SQL Execution) should only be authorized to administrators

Q: What if I want certain AI employees to bypass user permission restrictions?

A: Use workflow custom business tools:

• Create workflows to implement specific business query logic
• Control data scope and access rules in workflows
• Configure tools for AI employees to use
• Control who can call this capability through AI employee access permissions

Q: What's the difference between Overall Analytics and SQL Execution?

A:

Best Practices

1. Default to User Permissions: Unless there is a clear business need, prioritize using system built-in tools that follow user permissions
2. Templated Standard Analysis: For common analysis scenarios, use the Overall Analytics pattern to provide standardized capabilities
3. Strictly Control Advanced Tools: High-privilege tools like SQL Execution should only be authorized to a few administrators
4. Task-level Isolation: Configure sensitive tasks in specific blocks and implement isolation through page access permissions
5. Audit and Monitoring: Record AI employee data access behavior and regularly review abnormal operations