Show HN: SuperLocalMemory – Features Jev/Laya as Judges on Every Recall

Category: ai-ml

Tags: ai-memory, mcp, local-first, privacy, multimodal-search, vector-database

Score: 8.0/10 (Innovation: 8, Technical: 8, Documentation: 8, Utility: 8)

SuperLocalMemory is a governed memory system for AI agents that runs entirely on your machine, supporting multiple agents through MCP. It distinguishes itself with a three-tier recall system that refuses to bluff, on-device judge models (Laya and Jev) to verify answers, and multimodal search of images/PDFs via Google's EmbeddingGemma 2. Backed by four arXiv papers, it offers a local, privacy-preserving alternative to vector stores that always return something.

Target audience: AI engineers, developers building agents, and privacy-conscious teams

Repository: https://github.com/qualixar/superlocalmemory · Python · AGPL-3.0 · 254 stars

View on Hacker News