MemShield Zero-Trust AI Memory Suite
Interactive Prototype · Cloud Run + AlloyDB Express (PG 18.4)

See a cross-tenant AI memory leak happen live — and watch PostgreSQL stop it in the kernel.

An LLM never touches a database; it only answers from the memories your agent puts into the prompt. Run S5 below to strip the SQL WHERE clause completely on both sides, then inspect continuous hourly alerts and DPDP reports in Monitor or review the system architecture.

Step 3: Architecture →
adk-leak-arena v0.1.0 Unprotected pgvector vs adk-memshield (PostgreSQL FORCE RLS)
DB: probing... Ollama: probing... Gemini: probing...
Targets: unprotected, memshield
Hardware Load Notice: High-load stress parameters or local LLM inference selected.
Use [KILL RUN] inside the benchmark modal to abort and restore the 36-row baseline immediately.
RLS Write-Policy & CTF Lure Injector (/api/inject) Test owner spoofing, clearance write-up escalation, and private/shared lure writes against PostgreSQL WITH CHECK
Performance & RLS Kernel Overhead Unprotected SQL:- MemShield RLS SQL:- RLS Kernel Delta:- End-to-End Agent Turn:- Stage a scenario, then click Run Side-by-Side Query.