I built a persistent memory layer for Claude: here's a simple way to test it

Claude is useful until you open a new chat and have to explain your project all over again. So I built Cortex: an external semantic memory that Claude can access through MCP. Claude helped build it. Free to try for 30 days, no card; paid plans after that. The simplest test is this: In one chat, ask Claude to save a real project decision, including why you made it. Close the chat and delete it, so the answer cannot come from your chat history. Open a completely new one. Ask a question that depends on that decision. Without external memory, the new chat has no idea what you are referring to. With Cortex connected, Claude can retrieve the decision and its reasoning by meaning, even if you phrase the question differently. Cortex sits outside the individual conversation, so the memory does not disappear when the chat ends. It is also not tied to Claude itself: the same memory layer can be used by other compatible AI clients. I'm the developer, so yes, this is self-promotion. But the test is real and takes a few minutes. Claude setup: skynetlab-cortex.com/claude If you try it, I'm interested in where it works, where it fails, and what you would actually use persistent AI memory for.