Memoars: Encrypted Shared Memory Layer for AI Assistants
Dependent-Leg-3429 · reddit · 2026-08-07
Memoars is a tool designed to solve the pain point of context and memory sharing among multiple AI assistants. It allows you to connect a unified memory bank to your daily AI models via MCP (Model Context Protocol).
Key Features:
- Privacy & Security: Memory content is encrypted locally using XChaCha20-Poly1305 and written directly to storage you own (R2, S3, local fs, etc.), avoiding vendor lock-in.
- Conflict Resolution: The coordinator only handles metadata and cannot read the content. All changes land in an append-only, hash-chained log to prevent silent write conflicts and allow version tracing.
- Permission Isolation: Uses an Orgs → Workspaces → Identities permission model, with each workspace having its own passphrase to enforce isolation.
The product is currently invite-only, and the client is being open-sourced soon.
More from coding & agent
- NVIDIA to Showcase GR00T and ROS 2 Robot Manipulation in Live Demo — chrismatthieu · 2026-08-07
- resume-from: Open-Source Tool for Seamless Session Handoffs Between AI Coding Agents — alexei_led · 2026-08-07
- Claude Code Weekly Downloads Jump 6x to 3 Million in a Month, Hooking Tech Industry — ns123abc · 2026-08-07
- Yahoo Finance MCP: Real-Time Stock and Financial Data for AI Assistants — modelcontextprotocol · 2026-08-07
- Xcatcher: Open-Source MCP Tool to Fetch X Posts by Handle — modelcontextprotocol · 2026-08-07
- Building an RTS Game with Claude 3.5 Sonnet: Complex Logic Fails — dscape · 2026-08-07