🤓 Latest Submissions
AgentGuard AI — Secure AI Before Deployment
AgentGuard AI is a pre-deployment security and governance review platform for enterprise AI agents, built primarily with native.builder during the hackathon. As companies give AI agents access to sensitive data, databases, email, CRM systems, and high-impact tools, the question is no longer only “Can this agent work?” but also “Is it safe to deploy?” AgentGuard answers that question before production. The product combines two complementary layers. First, a deterministic risk engine calculates a reproducible 0–100 score across five dimensions: Sensitive Data Exposure, Tool & Action Privilege, Prompt Injection Exposure, Human Oversight, and Auditability. The score is rule-based, so an LLM cannot arbitrarily change it. Second, DeepSeek runs through a Supabase Edge Function as a contextual reasoning layer, generating realistic attack scenarios, contextual findings, and prioritized governance recommendations based on the submitted agent configuration. Each assessment is persisted in Supabase with anonymous authentication and row-level security. Teams can return to the dashboard, review historical assessments, and reopen completed reports without rerunning the AI analysis. The hackathon MVP demonstrates the complete workflow: configure an AI agent → run deterministic scoring → receive DeepSeek contextual analysis → generate governance recommendations → persist the assessment → revisit it from the dashboard. AgentGuard is intentionally scoped as a pre-deployment review system. It does not claim to provide real-time runtime blocking or enterprise IAM in the current MVP. Its value is to make AI-agent deployment more explainable, repeatable, and governable before risk reaches production.
10 Aug 2026
