
Cortex
Memory for AI assistants that decides what is worth keeping
I launched Cortex on Product Hunt on August 24. It is a hosted memory for AI assistants: you connect it to ChatGPT or Claude over MCP and both share the same memory of your decisions and context, with a filter that rejects about 80% of what clients try to store because it is redundant.
Numbers so far, in case they are useful to someone at the same stage.
Product Hunt: 84 upvotes, finished the day at #16. Good for morale, small in traffic.
AlternativeTo: listed on August 26. I also created the entries for the products people compare it with, so the "alternatives" links point both ways.
Trial accounts: over 100 in the first two weeks. When I last checked the split, at 83 accounts, 65 had come from the ChatGPT connector directory, which I submitted to almost as an afterthought. The website, the launch posts and everything else together produced the rest.
Paying customers: zero, on purpose. I opened trials before I had a company to invoice from; the company is being set up now and paid plans go live after that.
What I learned: for a developer-ish product, being inside the tool people already use beats being on a launch site. The connector listing has a tiny description and no screenshots, and it still outperformed everything else by a wide margin. Also: the quality gate rejecting 80% of writes is the feature people ask about most, and the one that confuses new users most. The response always includes the reason for a rejection, and that helped.
I am solo, self-taught, in Bergamo, Italy, no funding. Happy to answer anything about MCP distribution or memory design.
Cortex: https://skynetlab-cortex.com/?utm_source=indiehackers&utm_medium=post
About
Every AI assistant I used forgot my projects between sessions; the ones with a memory kept everything, noise included. Cortex is a hosted memory over MCP that decides what is worth keeping. Solo, self-funded, Bergamo.

Comment