Hey makers Launched YourMemory today, open-source, self-hostable memory for AI agents.
Every morning my agent treated me like a stranger: same context re-explained, same preferences forgotten, every session from zero. Most "memory" tools just bolt on a vector DB that hoards everything until retrieval drowns in noise, a goldfish with a bigger bowl.
So I built memory that works more like a brain:
Consolidates related facts into clean summaries, sharper over time, not bloated
Decays stale memories on an Ebbinghaus forgetting curve, but keeps the load-bearing stuff
Hash-chained audit trail so you can actually trust what's stored
Most AI memory tools (Mem0, Zep, vector DBs) are storage with search, they hoard everything and bloat into noise. YourMemory behaves like a brain:
🧠 It consolidates related facts into clean summaries, so memory gets sharper, not bigger (444 → 16).
♻️ It recovers your agent's context after every reset.
🔒 It proves every change with a tamper-evident audit trail.
👥 It shares role-based team memory pools.
Everyone else's store gets worse as it grows. Ours gets better.
I got tired of my AI agents (Claude/Cursor) losing the plot during long coding sessions because their context window was full of junk.
I spent my nights building YourMemory, a local MCP server that uses biological decay (Ebbinghaus curve) to prune stale logs while keeping important architectural "chains" alive.