Overview
Tool Manifest Validation guardrail monitors AI responses for tool calls and validates them against configured allowed or blocked tool lists. This prevents unauthorized tool usage and ensures AI agents only access approved tools.Configuration Options
Validation Mode
Choose how to validate tool calls:- Whitelist: Only allow tools explicitly listed in the allowed tools list
- Blacklist: Block only tools listed in the blocked tools list
Action on Violation
Specify what happens when a violation is detected:- Block: Completely block the request containing unauthorized tools
- Filter: Remove unauthorized tools from the response and allow the rest
- Audit: Log the violation but allow the request to proceed
Tool Lists
Configure which tools are allowed or blocked:- Allowed Tools: List of tool names that are permitted (for whitelist mode)
- Blocked Tools: List of tool names that are prohibited (for blacklist mode)
How It Works
The guardrail analyzes AI responses for tool calls and validates each tool against the configured lists:- Extract Tool Calls: Identifies all tool calls in the AI response
- Validate Against Lists: Checks each tool against allowed/blocked lists based on validation mode
- Apply Action: Takes the configured action (block/filter/audit) when violations are found
- Log Results: Records all validations and violations for monitoring
Use Cases
- Security: Prevent AI from accessing sensitive or dangerous tools
- Compliance: Ensure only approved tools are used in regulated environments
- Cost Control: Limit tool usage to prevent unexpected costs
- Access Control: Restrict tool access based on user permissions or roles
Best Practices
- Use whitelist mode for maximum security in production environments
- Regularly review and update tool lists as your system evolves
- Monitor violation logs to identify potential security issues
- Test configurations thoroughly before deployment
- Consider using audit mode initially to understand tool usage patterns