Verity: Open-Source Permission-Aware Memory for Multi-Tenant Agents
mattyboombalatti · reddit · 2026-08-04
A developer open-sourced Verity (Apache-2.0), a permission-aware memory system for multi-tenant agent/RAG stores, designed to architecturally prevent cross-tenant data leaks.
The Problem: When an agent generates a summary for Customer A in a shared memory store, the data typically loses its permission tags. This allows Customer B to accidentally retrieve A's sensitive data via standard semantic queries, with no system errors.
Flaws in Current Fixes:
- Namespaces/Metadata filters: A single untagged write creates an indefinite leak.
- System prompts: Easily bypassed, acting only as a suggestion rather than a boundary.
- Post-retrieval filtering: Kills recall and fails open if a code path skips the filter.
Verity's Approach:
- Index-level enforcement: Caller identity compiles into a mandatory pre-filter. Unauthorized rows are never candidates. No model involved, fails closed.
- Inherits ACLs automatically: Syncs and resolves complex nested group permissions directly from Google Workspace, SharePoint/Entra, and Salesforce. Revocations sync automatically.
Status: v0.1 with sync-based propagation (minutes of lag). Zero leaks in internal cross-tenant testing, though no external audit yet. Not intended for single-user memory setups.
More from coding & agent
- Karpathy Tests Claude Opus: Writes 5,500 Lines of Code in 2 Hours to Generate 3D Film — iamrobotbear · 2026-08-04
- Automating Agent Optimization: Devs Share Recursive Auto-Improvement Workflows — iamrobotbear · 2026-08-04
- Easy to Prototype, Hard to Maintain: OpenHands Shares Coding Agent Pitfalls — rajistics · 2026-08-04
- OpenBMB Launches Dual-Agent System to Automate Supercomputing Acceleration — aigclink · 2026-08-04
- Cloudflare Plans to Launch a Wallet Tailored for AI Agents — vista8 · 2026-08-04
- 76 Production-Ready Skills Released to Expand Claude's Capabilities — tom_doerr · 2026-08-04