The four-step memory loop, explained
claude-mem is the open-source engine; CMEM Cloud is the sync layer. Install it, let a second model take notes as you work, mirror your DB to the cloud behind one private MCP link, and recall everything from any agent on any machine.
From install to total recall in four steps
claude-mem is the open-source engine. CMEM Cloud is the cloud sync layer. Together your agents never start from zero again.
Recall that gets cheaper as it grows
Observations are written out-of-band, so capture never blocks your agent. Recall stays cheap because of progressive disclosure: a session index ~800 tokens for 50 observations surfaces just enough context to find the right memory, then fetches the detail on demand — up to 95% fewer tokens in our docs' worked example (18,000 → 970 tokens).
The store is built to scale with your work. In claude-mem's own corpus that is already 92k embeddings across 253 projects — local-first on SQLite, mirrored to CMEM Cloud only when you opt in, and reachable from every agent through one private MCP link.
Start syncing your memory
The engine is open source and free to install today. Cloud sync is $0 during early access — join the waitlist and your private MCP link arrives by email.