shodh-memory and mnemosyne
These two tools appear to be direct competitors, both aiming to provide a foundational, self-improving cognitive memory operating system for AI agents, with **shodh-memory** emphasizing a single, offline binary and **mnemosyne** highlighting multi-agent support and persistence.
About shodh-memory
varun29ankuS/shodh-memory
Cognitive memory for AI agents — learns from use, forgets what's irrelevant, strengthens what matters. Single binary, fully offline.
Implements local embeddings and Hebbian learning to achieve sub-200ms memory storage without LLM API calls, with automatic activation decay and spreading activation for relevance-based recall. Available as MCP server for Claude/Cursor, HTTP API, or native Rust/Python libraries; also supports robotics frameworks (ROS2/Zenoh) and includes a TUI dashboard for memory visualization and GTD task management.
About mnemosyne
28naem-del/mnemosyne
Cognitive Memory OS for AI Agents — persistent, self-improving, multi-agent memory
Implements a 5-layer cognitive architecture with algorithmic entity extraction, temporal knowledge graphs, and multi-signal retrieval scoring—running zero LLM calls during ingestion for $0 per memory stored. Supports real-time pub/sub broadcast across multi-agent meshes via Redis, bi-temporal tracking, and reinforcement learning that auto-promotes high-utility memories. Integrates with Qdrant for vector storage, optional Redis/FalkorDB for caching and graph queries, and works with any embedding provider.
Related comparisons
Scores updated daily from GitHub, PyPI, and npm data. How scores work