Player not loading? Watch on YouTube
This comparison examines claude-mem and agentmemory as open source memory tools for a coding assistant. Both aim to carry project context across sessions so users spend less time repeating decisions and debugging history.
The speaker describes claude-mem as a background plugin with five lifecycle hooks. It captures file edits and commands, compresses session activity into observations, and stores them locally for later sessions. Agentmemory uses four memory tiers: working, episodic, semantic and procedural. The speaker says frequently used memories strengthen while stale ones decay, and cites published retrieval accuracy of 95.2% on an academic benchmark associated with ICLR 2025. The video provides no comparable benchmark for claude-mem.
Setup follows repository installation instructions. Agentmemory also requires a memory server in a separate terminal, followed by plugin installation and a Claude Code restart. Both have web viewers; agentmemory's viewer includes a knowledge graph. The speaker presents persistent memory as useful for projects spanning days or weeks, with less benefit for brief, one-off questions.
Security is a substantial part of the comparison. The speaker cites claude-mem issue #1251 for network exposure and clear-text API key concerns, and describes agentmemory's secret filter and localhost default. Local storage alone does not establish that these tools are private. The advice is to keep services bound to localhost and check settings.json for exposed keys.
