AGENT SECURITY OPERATIONS
Security overview
Review tool-call decisions, test policies, and verify your audit trail.
START HERE
Try the security workflow
Replay a synthetic call, inspect its verdict, then verify the audit trail. No real tools are connected.
Backend status unavailable · judge mode unknown
Intercepted
0
LAST 100 EVENTS
Blocked
0
LAST 100 EVENTS
Needs review
0
LAST 100 EVENTS
Allowed
0
LAST 100 EVENTS
POLICY TESTING
Scenario simulator
PUBLIC DATA LAB
Evaluate labeled text samples
Test your public prompt-injection examples against the current deterministic context rules.
This public website sends the selected rows to its hosted backend. Evaluation uses deterministic rules only—no Nebius/Gemini calls—and the app does not add rows to the audit ledger. Only upload public, non-sensitive data; do not upload private, personal, or confidential information. Hosted platforms may process request data.
Upload JSON or JSONL (one JSON object per line) with
text and label fields. Label 1 means attack; 0 means benign. Optional category is accepted but not included in the report. Maximum 500 rows, 8,000 characters per row, and 2 MB per file.No file selected
DECISION PIPELINE
Latest trace
Run a scenario to inspect its policy decision and audit proof.
TAMPER-EVIDENT RECORD
Audit ledger
Chain integrity
Verify the SHA-256 event chain and Ed25519 signatures.