GhostRange

GhostRange is an autonomous cyber-defense proving ground built for safe agent execution. Instead of allowing an AI agent to experiment directly on production infrastructure, GhostRange reconstructs an authorized system inside isolated disposable environments on Vultr. From a repository, incident description, infrastructure configuration, and optional runtime evidence, GhostRange builds a living digital twin of the affected system. GhostDirector creates competing hypotheses and decides what experiments would best distinguish them, while GhostScheduler determines how those experiments should execute based on dependencies, worker availability, runtime, and cloud cost. GhostRange then creates temporary Vultr workers and experimental worlds, reproduces the incident, collects evidence, tests causal interventions, evaluates multiple remediation candidates, and actively searches for counterexamples that could invalidate a proposed fix. The safety boundary is enforced by GhostShield: models can propose typed actions, but they do not receive unrestricted Vultr credentials or direct infrastructure authority. Protected effects must pass deterministic authorization, resource-ownership checks, budget limits, and existing human-control boundaries. Every important result is connected to its source, experiment, worker, evidence, and decision history through GhostLedger. Once work is complete, temporary infrastructure is drained and destroyed while the evidence remains available for review. The result is not simply an AI-generated answer to “what went wrong?” GhostRange produces an experimentally tested explanation, rejected alternatives, verified-under-tested-conditions remediation evidence, execution timing, compute cost, provenance, and the limitations of what was actually tested.

Note: the demo URL is substituted with the GitHub repository link.