Hey Product Hunt! I’m part of the engineering team building CordonData. When we started building internal AI tools, we hit a massive security wall: standard RAG pipelines typically ignore existing Access Control Lists (ACLs). If an employee queries an LLM, there is a very real risk the model retrieves and summarizes a sensitive financial or HR document they shouldn't have access to. To solve this, we spent the last few months building an engine that actually respects source permissions. Instead of copying massive raw files and creating a nightmare of data duplication, we extract content from your existing corporate silos directly into clean, lightweight Markdown. We run a dual-engine hybrid retrieval, but the core innovation is that we enforce Document-Level Security (DLS). The system inherits permissions directly at the document level. At query time, the LLM only receives context from the Markdown documents the executing user is explicitly authorized to view. Because enterprise data is sensitive, the entire system is designed to be air-gapped and deployed on-premise, complete with universal audit logging. For teams that don't have a centralized file repository, we also built a secure internal drive directly into the frontend to manage and permission files natively. If you are an Enterprise Architect, Data Lead, or just dealing with the exact headache of keeping LLMs secure behind a firewall, we’d love for you to check it out. We would love any brutal feedback on the architecture or our approach to handling Document-Level Security!

CordonData

Air-gapped RAG engine & secure drive with native DLS & Audit

Air-gapped RAG engine & secure drive with native DLS & Audit