Shield
See what your AI is doing and protect the data moving through it.
Overview
Shield brings visibility and protection to your AI workflows. Inspect requests, follow agents across models, tools, and apps, and see where sensitive data is redacted or an action is blocked by your policies.
Explore Logs and Traces
Use Logs to review individual requests and their protection outcomes. Switch to Traces to follow a request through its model and tool calls. Filter by model, activity, time window, or outcome to find the request you want to investigate.
Follow a Request
Open the Visual view to see how agents hand work to each other. Select a step to inspect its payload and protection decision. The graph shows allowed, redacted, and blocked steps, while the execution timeline makes parallel activity and timing easier to follow.
- Choose a request from the activity list.
- Follow its branches across models, tools, and apps.
- Inspect the step where data was masked or an action was stopped.
- Use trace replay to review the sequence and timing of the workflow.
Redact Sensitive Text
Send text to the redaction API, choose a detection model, and specify the labels you want to detect. Choose how detected values should be replaced before the text moves to another system.
Choose a Detection Model
Swift uses predefined labels for common types of sensitive data. Forge lets you describe the concepts you want to detect. The API reference includes request examples, supported labels, and transform modes.
Review Protection Decisions
A redacted step shows where sensitive values were masked. A blocked step shows an action stopped before execution. Review the payload and decision together to understand how your policies apply across the workflow.