YourMemory

About YourMemory

YourMemory is an AI memory manager for agent workflows that applies a biologically inspired decay model to keep context focused and relevant. Launched this week, it aims to reduce token waste by pruning stale memories while reinforcing items that matter.

Review

YourMemory combines an Ebbinghaus-style forgetting curve with a graph engine to manage agent context. The vendor reports an 84% reduction in token waste and a 52% Recall@5 on the LoCoMo benchmark, and the system is implemented as a local-first MCP storage solution.

Key Features

• Self-pruning MCP memory using an Ebbinghaus decay model to age out low-value context.
• v1.3.0 Graph Engine that retrieves related "neighbour" memories beyond simple keyword matches.
• Reported -84% token waste and 52% Recall@5 (LoCoMo benchmark) for leaner context and improved retrieval.
• 100% local-first storage (DuckDB) with an open-source launch offering and zero infra requirements.

Pricing and Value

The product is listed as free and open source at launch, which lowers the entry barrier for developers and teams experimenting with agent memory management. Value comes from reduced token consumption (which can lower runtime costs with hosted LLMs), improved recall for agent reasoning, and the ability to run everything locally without additional infrastructure.

Pros

• Significantly reduces token waste, which can improve prompt efficiency and lower costs.
• Graph-backed retrieval helps surface related, contextually relevant memories that simple vector search can miss.
• Local-first approach (DuckDB) and open-source availability make it attractive for privacy-conscious workflows and experimentation.
• Configurable importance and chain-aware pruning help protect low-frequency but high-importance items.

Cons

• As an early launch, integrations and developer ergonomics may require additional polish and documentation.
• Edge cases exist where long-dormant but load-bearing decisions need manual linking or importance tuning to avoid accidental pruning.
• Teams using hosted multi-service pipelines may need to adapt their architectures to a local-first storage model.

Overall, YourMemory is most useful for developers and teams building long-lived agent workflows or applications that suffer from context rot and high token usage. It fits especially well where local control, cost-consciousness, and improved agent recall are priorities.

Open 'YourMemory' Website