SuperLocalMemory 4.0: A Governed Memory OS for AI Agents
A new paper on arXiv (arXiv:2608.08253) has unveiled SuperLocalMemory 4.0, a memory operating system designed for AI agents that prioritizes local governance. This innovative system employs various retrieval techniques, including dense semantic, BM25 lexical, temporal, Hopfield-associative, and spreading-activation, all unified through reciprocal-rank fusion. Key features encompass a governed learning and behavior layer, bi-temporal recall, multi-scope memory (personal, shared, global), role-based access control, GDPR-compliant export and verified erasure, audit trails, and a checklist for the EU AI Act relevant to deployment contexts. The latest version also introduces a reliability spine for the primary write path, ensuring secure memory transactions and accessible runtime through CLI, MCP, an HTTP daemon, and a dashboard.
Key facts
- SuperLocalMemory 4.0 is a governed, local-first memory operating system for AI agents.
- It combines dense semantic, BM25 lexical, temporal, Hopfield-associative, and spreading-activation retrieval via reciprocal-rank fusion.
- The system includes a governed learning and behaviour layer, bi-temporal recall, and multi-scope memory (personal, shared, global).
- It features role-based access control, GDPR-oriented export and verified erasure, audit trails, and an EU AI Act checklist.
- Version 4.0 introduces a reliability spine with generation-fenced admission, a policy registry, and verifiable memory transactions.
- Memory transactions include per-projection apply, verify, compensate, and erase owners, plus hash-checkable completion manifests.
- The runtime is available through CLI, MCP, an HTTP daemon, and a dashboard.
- The paper is available on arXiv with identifier 2608.08253.
Entities
—