Hound: Open AI Harness for Security Audits

Hound is an honest take on AI security research: orchestration can expand coverage, but verification remains the hard part.

• –Builds adaptive knowledge graphs to reason across architecture, access control, data flows, and code relationships
• –Uses multiple specialized passes before separate judging stages review candidate findings
• –Supports focused investigations and final confidence scoring rather than treating every model output as a vulnerability
• –Cloud-only operation limits use on confidential code, since scans send source to external model APIs
• –Best suited today as a first-pass audit assistant for small-to-medium codebases, not an autonomous replacement for security researchers