Launching today
Neutrally
One memory for Claude, ChatGPT, Cursor and any MCP client
5 followers
One memory for Claude, ChatGPT, Cursor and any MCP client
5 followers
Your memory, across every AI you use. Save it once in Claude, pick it up in ChatGPT, Cursor, Grok or any MCP client. Saving and recall are plain language, just ask. Built for devs and teams tired of re-explaining their project to every new chat. Scores 89.4% on LongMemEval-S. Privacy-first: scoped to your account, never trained on, with on-prem for teams. Claude Partner Network member, one-click setup. Watch the demo to see it work across models. Free to start.
Hey Product Hunt, Ali here.
I built Neutrally because context never really followed me. I'd keep claude-md files and re-paste the same background into each tool, but it was always partial. Something important would get left out, and even staying inside Claude, detail would drop between sessions and I'd be rebuilding what I'd already explained. Plan in ChatGPT, move to Claude to build it, and half the thread didn't make the trip. If you're a dev or on a team, you know the tax: the manual workarounds help, but they never carry everything that mattered.
So I made the thing I wanted: a memory layer that sits above the models, not inside any one of them. You connect it once, and the same memory is there whether you're in Claude on your phone, ChatGPT on your laptop, or Cursor in your terminal.
A few honest notes:
It works with any client that supports MCP connectors. I've tested Claude, ChatGPT, Cursor and Grok, and because it's built on the open MCP standard, any compliant client works the same way. If your client speaks MCP, you're covered.
Saving and recall are plain language. Say "save this to Neutrally" to keep something, and ask for it back the same way.
If you live in Claude Code like I do, there's a plugin that auto-loads your memory at the start of each session and captures every turn automatically. No commands, it just runs.
It scores 89.4% on LongMemEval-S, a strong result on a recognised memory benchmark. Happy to talk methodology with anyone who cares about the details.
Privacy is the point. Your memory is scoped to your account, never used to train models, and you can export or delete it anytime. On-prem available for teams who need everything in-house.
I'm a Claude Partner Network member, and MCP setup is one click.
Free tier to start. Pro is £99/year.
I'm here all day answering everything. If you try it and something feels off, tell me. That's exactly what I want from today.
Two things I'd love your take on: where would portable memory save you the most time, and what would make you trust a third-party memory layer with your context?