Second Brain for AI
Persistent memory for Claude, ChatGPT & Cursor. Free.
Start new thread

Second Brain for Teams - Give your team one AI memory. Keep your own private.

by
•
Second Brain gives your team one AI memory that works across Claude, ChatGPT, Cursor and any MCP client. Personal memories stay private, shared ones reach everyone, and every write is logged. Free, MIT licensed, running in your own Cloudflare account.

Add a comment

Replies

Best
Maker
📌
Hey Product Hunt, I’m Rahil, the maker of Second Brain. Second Brain has gone through a few versions now. First, it was a way to give all your AI tools the same memory. Then I added a graph that connects those memories automatically. After that came the desktop app, mostly because I realized “just deploy this to Cloudflare” was not a reasonable onboarding experience for most people. This is the next thing I’ve wanted to build for a while: Teams. Someone once described using several AI tools as having five tiny brains that all confidently remember different things. That gets even worse when you add a team. Everyone has their own AI tools, their own context, and their own version of why a decision was made. You end up explaining the same things over and over again. A five-person startup asked me a pretty simple question: can we all use the same Second Brain? At the time, the answer was no. The problem was deeper than adding team accounts. One Second Brain deployment had one memory, and anyone with the token could access everything in it. That works when the only person using it is you. It obviously doesn’t work for a company. So I didn’t want Teams to just be a permissions screen on top of the existing architecture. Now every member has their own private memory, and the team has a shared memory. When you recall something, Second Brain can search across both. When you save something, it stays private by default. And when something should belong to the team, you can just say so: “Remember this and share it with the team.” You can do that from Claude, Cursor, or whatever MCP client you already use. There’s no separate knowledge-management workflow to remember. There are a couple of things I deliberately haven’t solved yet. There are no project-level memory layers. It’s personal or team-wide for now. And converting a personal brain into a team brain is one-way. There’s no downgrade button. Once other people have memories in that shared workspace, turning it back into a personal brain could mean deleting data that belongs to them. I’d rather not ship that button than ship one that can do something dangerous. Second Brain is still MIT licensed. It still runs entirely in your own Cloudflare account, and there’s still nothing to pay me. There is a real infrastructure cost for Teams, though. A personal Second Brain can comfortably live inside Cloudflare’s free tier. A team deployment probably won’t, so if you’re setting one up, I’d budget for Cloudflare’s paid Workers plan. If you work with a team, I’m curious: what’s the one thing you’re tired of having to explain to your AI over and over again? I’ll be around all day. Code: github.com/rahilp/second-brain-cloudflare
Maker

The team stuff got the headline. Here is what else went into v3.

Your memories are better at knowing how they relate to each other. Before, everything was connected to everything in the vaguest possible way, which is another way of saying nothing was really connected. Now a memory can know that one thing came after another, or caused it, or was the decision that settled it. It also stopped connecting near-identical notes to each other and calling that an insight.

Setup got a proper going over. We had people who were not me try to install it and write down every place it annoyed them, which was humbling and useful in that order. It now checks it is pointing at the right account before it starts making changes, warns you properly before anything irreversible, and works in Italian.

If someone invites you to a team, you get an actual product rather than a wall of "admins only". You can see who wrote what and when. Invites work without you explaining them in Slack afterwards.

Code:

great product, congrats on the launch.

the private-by-default, share-on-request model is the right instinct, but I'm not sold on the edges yet. what happens to someone's private memory when they leave the team - deleted, orphaned, or does it need an explicit handoff? and since "remember this and share with the team" is just a phrase decided in the moment with no separate review step, what stops someone from assuming something was shared when it wasn't, once there's no UI to double check after the fact?

Maker

 Yep, we handle both of these explicitly.

When someone is removed, their private workspace is deleted with them. Anything they already shared stays in the team layer, with the original attribution preserved in the audit trail. There’s also a separate suspend option that cuts access without deleting their data.

And private vs shared isn’t just something you have to infer from the conversation. The UI has explicit private/shared layers, shared-state indicators and controls to share or make memories private again. The graph uses the same layer distinction, and admins have an activity/audit view. Underneath that, the boundary is enforced at the query layer, so even an admin can’t recall, list or search another member’s private memories.