AgentGuard live demo

Model-agnostic security harness for AI agents — every tool call below is evaluated by the real deny-by-default policy engine and written to the real append-only audit ledger. No mocks, no simulated decisions.

Demo sandbox — policy decisions only, nothing is executed. Hosted on Cloudflare Workers.

Try: read_file on welcome.txt (allowed) vs /etc/passwd or ../../etc/passwd (denied: path escapes sandbox), run_command on anything (denied), or scan_text on text containing AKIAIOSFODNN7EXAMPLE (flagged).

Audit ledger (live tail)

total 3 · allowed 1 · denied 2

tstoolargsdecisionreason
1791821415scan_text{'chars': 30, 'findings': ['aws_access_key']}allowallowed: token ff140a2268a5fe02 covers scan_text
1791821412read_file{'target': '../../etc/passwd'}denypath escapes sandbox
1791821410read_file{'target': '/etc/passwd'}denypath escapes sandbox