Stiletto is an AI red-teaming agent that runs a full pentest campaign (recon, exploitation, post-exploitation, lateral movement, persistence, and a ranked report) with structural safety guarantees: the model reasons and returns validated JSON while a LangGraph state machine owns scope and execution behind a fail-closed guard, and every escalation clears a real human-approval gate the model cannot skip. The operator runs the whole campaign from a live web console that streams tool activity in real time, which is where each of those approvals happens.
The source is private for now, but I'm happy to give a live demo on request.
Request a demo: use the contact form at https://nopoz.dev and mention stiletto.