Show HN: AI memory with biological decay (52% recall)

YourMemory is a persistent memory layer for AI agents. Like human memory, it factors in importance, recall count, and elapsed time so that memories decay, and old or weak entries are automatically cleaned up.

The core benchmark is LoCoMo-10. It was evaluated on 1,534 QA items and 10 multi-session conversations, and it reportedly achieved Recall@5 59%. It states that this is higher compared to Zep Cloud 28% on the same benchmark.

The implementation is local-first.

• Supports Python 3.11~3.14
• Get started with pip install yourmemory and yourmemory-setup
• Runs on DuckDB with no separate Docker, DB, or external services needed
• Uses NetworkX, sentence-transformers, spaCy, and APScheduler

Memory decay is calculated using a modified Ebbinghaus curve.

• effective_λ = base_λ × (1 - importance × 0.8)
• strength = importance × e^(−effective_λ × days) × (1 + recall_count × 0.2)
• score = cosine_similarity × strength

Search happens in two stages.

• Round 1: vector search finds relevant memories.
• Round 2: graph BFS expands to related memories.
• Connections with cosine_similarity ≥ 0.4 are used as graph edges.

Memories are divided into categories such as strategy, fact, assumption, and failure, each with different retention periods. It's designed to attach to MCP-compatible clients and automatically call recall_memory, store_memory, and update_memory.

It provides setup examples for Claude Code, Claude Desktop, Cursor, Cline, OpenCode, and more, and also supports a structure where multiple agents share both shared and personal memory.