Ori Mnemos

About

Ori Mnemos serves as a foundational memory layer for AI agents, addressing the common problem of agents losing context and identity between sessions. Unlike traditional databases or vector stores, it uses plain markdown files on disk for knowledge, identity, and operational data, enabling human-readable, version-controlled memory without external dependencies. It ensures agents wake up with prior knowledge, goals, and a consistent identity, facilitating continuous learning and decision-making by acting as the crucial link between an agent's short-term context window and its long-term knowledge base.

Key Features

• Markdown-Native Knowledge Graph: Builds an in-memory graph from wiki-links in plain markdown files, supporting structural queries and community detection.
• Model Context Protocol (MCP) Integration: Exposes a comprehensive set of memory operations via JSON-RPC over stdio for any MCP-compatible AI agent runtime.
• Vitality Model: Explicitly models note decay based on access, connectivity, and metabolic rates, surfacing fading knowledge for review.
• 1 GitHub stars
• Agent Identity Layer: Automatically injects identity, goals, and methodology into agents, enabling continuity across sessions and deployments.
• Three-Signal Retrieval Engine: Fuses semantic, keyword (BM25), and graph (PageRank) signals for highly relevant information retrieval.

Use Cases

• Enabling AI agents to build, retrieve, and reason over a structured knowledge base without external database dependencies.
• Providing AI agents with long-term memory and consistent identity across multiple sessions and deployments.
• Facilitating continuous learning and state management for autonomous AI workflows.