Column-Level Permissions

Control data access at the column level. Hide sensitive information from specific agents while maintaining usability.

🔒 What Are Column Permissions?

Column permissions let you:

• Hide PII: Protect personally identifiable information
• Secure sensitive data: Hide salary, financial, or medical data
• Compliance: Meet GDPR, HIPAA, or other regulatory requirements
• Role-based access: Different users see different columns

🎯 Use Cases

⚙️ How It Works

Agent-Level Permissions

Column permissions are configured per agent:

1. Agent is created with access to specific tables
2. For each table, admin selects visible columns
3. AI cannot see or query hidden columns
4. Hidden columns are excluded from schema context

📝 Configuring Column Permissions

During Agent Creation

1. Navigate to Agents → New Agent
2. Select data source and tables
3. For each table, click "Configure Columns"
4. Select which columns to show
5. Save agent

Example: customers Table

✅ Visible Columns:
  - customer_id
  - customer_name
  - company_name
  - industry
  - region
  - created_at

❌ Hidden Columns:
  - ssn
  - credit_card_number
  - bank_account
  - internal_notes
  - credit_score

Editing Existing Agent

1. Navigate to Agents
2. Click on the agent
3. Click "Edit"
4. Go to "Table & Column Permissions"
5. Modify column selections
6. Save changes

🔍 Query Behavior

User Query:

"Show me all customer information for ID 12345"

Generated SQL (with permissions):

SELECT 
  customer_id,
  customer_name,
  company_name,
  industry,
  region,
  created_at
FROM customers
WHERE customer_id = 12345;

🎓 Common Patterns

1. Sales Agent (Limited PII)

Table: customers
Visible:
  ✅ customer_id, name, company, industry
  ✅ email, phone
  ✅ sales_rep_id, account_status
Hidden:
  ❌ ssn, credit_card, bank_account
  ❌ internal_notes, credit_score

2. Finance Agent (No PII)

Table: orders
Visible:
  ✅ order_id, customer_id, order_date
  ✅ total_amount, payment_method
  ✅ order_status
Hidden:
  ❌ customer_name, email, phone
  ❌ shipping_address
  ❌ credit_card_last4

3. Analytics Agent (Aggregates Only)

Table: employees
Visible:
  ✅ employee_id, department, job_title
  ✅ hire_date, employment_status
Hidden:
  ❌ first_name, last_name, email
  ❌ ssn, salary, bonus
  ❌ performance_rating, manager_notes

4. Executive Dashboard (High-Level Only)

Table: revenue
Visible:
  ✅ date, region, product_category
  ✅ total_revenue, order_count
Hidden:
  ❌ customer_id, sales_rep_id
  ❌ individual_order_details
  ❌ cost_of_goods_sold, profit_margin

🔐 Security Best Practices

🌐 Compliance Support

GDPR

• Right to privacy: Hide personal data from analytics
• Data minimization: Only expose necessary columns
• Purpose limitation: Different agents for different purposes

HIPAA

• PHI protection: Hide protected health information
• Minimum necessary: Only show data needed for job function
• Access controls: Role-based column access

PCI DSS

• Cardholder data: Never expose full credit card numbers
• Sensitive authentication data: Hide CVV, PINs
• Tokenization: Use customer_id instead of payment details

⚡ Advanced Features

Dynamic Masking

Show partial data instead of hiding completely:

credit_card_number → ****-****-****-1234
ssn → ***-**-1234
email → j***@example.com

Conditional Permissions

Show/hide columns based on conditions:

• Time-based: Hide salary until fiscal year end
• User-based: Managers see more than analysts
• Data-based: Hide columns for specific customer segments

🔍 Auditing & Monitoring

Access Logs

Track which users query which columns:

• User ID and timestamp
• Agent used
• Tables and columns accessed
• Query text and results

Permission Changes

Audit trail for permission modifications:

• Who made the change
• When it was made
• What changed (before/after)
• Reason for change

Compliance Reports

Generate reports for auditors:

• Column access summary
• Sensitive data exposure
• Permission changes over time
• Users with access to PII

🎯 Implementation Checklist

Initial Setup

1. ✅ Identify sensitive columns in all tables
2. ✅ Document column classification (public, internal, confidential, restricted)
3. ✅ Create agents with appropriate permissions
4. ✅ Test queries to verify hidden columns are inaccessible
5. ✅ Train users on data access policies

Ongoing Maintenance

1. ✅ Review permissions quarterly
2. ✅ Update as schema changes
3. ✅ Audit access logs monthly
4. ✅ Revoke unnecessary access
5. ✅ Document all permission changes

⚠️ Common Mistakes

🔧 Troubleshooting

Query Returns Unexpected Results

Cause: Hidden columns affecting joins or filters

Solution: Review which columns the agent can see. Add necessary columns for proper query execution.

User Can't Access Needed Data

Cause: Column hidden from their agent

Solution: Create new agent with broader permissions or assign different agent.

Sensitive Data Still Showing

Cause: User switched to different agent with more permissions

Solution: Review agent assignments and ensure proper column hiding.

📊 Example Configurations

Retail Company

Agents:
1. Sales Agent
   - See: customer contact info, order history
   - Hide: payment info, internal notes

2. Marketing Agent
   - See: demographics, purchase patterns
   - Hide: PII, payment info

3. Finance Agent
   - See: order amounts, payment status
   - Hide: customer names, contact info

4. Executive Agent
   - See: high-level metrics only
   - Hide: all individual customer data

Healthcare Provider

Agents:
1. Operations Agent
   - See: appointment times, facility usage
   - Hide: patient names, diagnoses

2. Billing Agent
   - See: patient ID, billing codes, amounts
   - Hide: medical details, doctor notes

3. Research Agent
   - See: de-identified medical data
   - Hide: all PII (names, addresses, DOB)

4. Admin Agent
   - See: all data (restricted to compliance team)

🚀 Next Steps

• Create an agent with column permissions
• Combine with role-based access control
• Column permissions API reference