Your Dropbox for agent knowledge (it works without you having to "work" it or maintain it.) Every agent you use reads and writes to the same brain, and so do your teammates and theirs. Start a new chat and it can know what you've all already worked on. Your IP, research, sci-fi romance novel or notes are encrypted and I won't train on it. Export it to markdown and leave whenever, because I think you shouldn't have your knowledge locked in anywhere, including OzBrain.
What's great
OzBrain is the third agent memory product that I've used and the one that I ultimately stuck with. The nice thing about OzBrain is that memory doesn't have to be overcomplicated. I find that simple is normally better when it comes to coding agents.
I just have instructions in my AGENTS.md that tell it to persist any decisions that were made in any design docs. From there, it has just continued to write and update any decisions made. It's very useful for iterating and experimenting, especially on AI pipelines where I'm tweaking the inputs and prompts a lot, then benchmarking against some output.
Memory should just be used when necessary and shouldn't drown the agent context. I find OzBrain with very simple setup already succeeds in both. I have not opened the UI. It just works.
What needs improvement
Honestly, simple is better, so I feel that it doesn't need to be improved drastically. I'm already close to 20 articles, and I somehow find that Codex continues to find the correct article without me having to point it towards.
vs Alternatives
Supermemory
NiaI think that the other tools in this space try to do some fancy stuff like indexing your code. These often produce graphs that are left stale. And I also find that they bloat context but don't provide as much value. On the other hand, OzBrain is much simpler in that I just need to tell it to remember something. It doesn't need to remember the technical implementation. I just use it as more of a fact ledger for things that I've tried before and decisions that I want to be committed to memory so that I don't fall into the same trap again. This is good enough to save me from having to check old git commits for much less tokens.
Ratings
Ease of use
Reliability
Value for money
Customization
Report
14 views
What's great
We have a lot of different surfaces for my start up, and our whole team had individual patched their own silo'd solution together with their agent, github, slack, granola etc. Most founders I chat with I ask "what are you doing for a team brain/data lake?" and most have some sub optimal, stuck-together fix. So I've been looking for this exact thing, Ozbrain, for months. And my review after two weeks using it daily:
it does exactly what I need, it's slotted sleamlessly into my daily work flow, it's great. The versioning is really cool too, we take it for granted with code and git, but having it for company wide knowledge/data is great. Make so much sense.
It's truely MCP first, set up was a prompt and a few follow up questions in my agent, super simple.
What needs improvement
I'd like a slack connector to run ozbrain queries in slack (like Claude Tag)
vs Alternatives
Our team had been using a frankenstein combo of github, google drive and shared skills. Ozbrain is way way better.
Report
10 views
@bubsexcited to see this one live! I've been on the beta for a few weeks, it's one of those immediately useful/straight into my daily workflow tools like Wispr flow/granola/claude routines. set up was a prompt or two, and I just added in a line to my common skills to update ozbrain when the skill runs.
I've found it super useful for product analytics and GTM insights, even kicking off a new feature. Feels like the glue between a lot of systems we run on nitrosend, I don't think I've been back to the UI since I set it up but I use it all the time (from my agent). Very cool
@gthartley Thanks George! Yup, while there is a UI, I think we're all moving Agent-First and that is where I expect 99.99% of usage happens for OzBrain. Glad it's going well for you!
@bubs nice launch! the idea of the having one shared brain across claude, cursor and other agent make a lot of sense, and also the Dropbox for agent knowledge likeness make the idea pretty easy to understand.
how does OzBrain decide when the something should be obsolete or kept as the current versions??
@amrita_prajapati2 Thanks! A lot of the "thinking" the brain does around current versions is handled by the system which has versioning built in, so every change is staged/promoted... and your agent that is making the most recent update is holding the context of where-else the content of the article is is working with is linked to/from. So it's really about create a system that helps your agents ensure they are working from the latest thinking.
@bubs congrats for launch🙌qq how does this compare to just using a shared Notion database with an API hook?
@priya_kushwaha1 Thanks! That's basically where I started... tables in my own DB, agents hooked in over MCP. It works until two agents write at once and clobber each other, or it rots because nobody maintains it, or the agent dumps the whole table into context instead of the part it needed. OzBrain is that same idea with those parts solved: stage/promote conflict handling with a log of who changed what, deprecate-don't-overwrite versioning, and structure so agents pull the right thing. Solo with one agent, a Notion hook is honestly fine. It's the multi-agent, multi-person case where the glue turns into a maintenance job.
Two agents proposing conflicting writes at the same time is the part I'd want to see. Who decides which one lands?
@justin_rockmore The agents who are trying to do the write would. It would surface the conflict to one and let them know about the collision. Both writes are preserved in the version history too so a human in the loop can always make an override decision. But I wanted the system to be able to automatically resolve issues vs stopping and surfacing for a human… at the pace agents work that is fast growing backlog for a person to have to manage.
OzBrain has been the easiest way to keep shared context between me (using Cursor) and my cofounder (using Claude Code). Incredibly easy to get started, and excited to see it launch!
How does this different from me using Notion ?
User friendly dropbox for context is a huge gap in the market!
If you are non technical and use multiple AI products Ozbrain is a no brainer.
"every agent and every teammate reads and writes to the same brain" is the part I'd want to poke at before adopting this for a team. if my research notes and a teammate's HR draft both live in the same shared brain, is there any scoping between people/projects, or does "shared" really mean flat and anyone's agent can surface anyone else's stuff? export-anytime is a good trust signal, but access control feels like the harder problem here.
@galdayan It's free to try. Would love to have you give it a go and happy to work quickly to adapt to improve where your team needs. You can share collections so it's nto a wholesale sharing of everything you do. Maybe there's a sales-team collection where the sales team reads/writes to, but the broader team has read access only.
@bubs the "maybe there's a sales-team collection" phrasing is what caught my eye - sounds like that's a pattern people can build with collections rather than something scoped and enforced by the product itself. is access control per-collection actually enforced today, or is it more of a convention teams have to agree to follow?