VeilMind prototype isolates private AI memories while sharing generalized lessons
Logical_Leading8882 · reddit · 2026-09-29
A developer built VeilMind, a prototype tackling a multi-tenant problem: when an AI consultant serves multiple clients, not sharing memory means starting from scratch, but sharing risks leaking client info.
The approach separates private experience from shared lessons:
- Each client gets isolated memory; on engagement end, a generalized lesson is extracted, checked for identifying info, and only then added to a shared playbook
- Includes an attack-testing harness probing whether the AI can be tricked into revealing another engagement's info
- Conflicting knowledge (e.g., phased migration worked for one client, failed for another) is surfaced rather than treated as universal truth
Positioned as a hackathon prototype, not a production security system; the author invites feedback from RAG, agent-memory and multi-tenant folks.
More from coding & agent
- Garry Tan Backs AI Browsers Like AsideAI: Agent Password Management Is the Key Layer — garrytan · 2026-09-29
- Open-Source Code Review Agent Uses Hindsight to Remember Team Decisions — aryanplayz630 · 2026-09-29
- Meeting Prep Agent Separates App State in SQLite from Long-Term Memory — Datrika_Nandhini · 2026-09-29
- Chestnut unveils 18-DOF Aero Hand plus matching exoskeleton for zero-gap humanoid data capture — chris_j_paxton · 2026-09-29
- RubyLLM hits 1.1M downloads in a month, 13M total — kieranklaassen · 2026-09-29
- Tutorial: Building a support agent with customer memory using Hindsight — Tricky-Coconut9833 · 2026-09-29