Hi I'm Sourav. I am building Memoair, the memory and context layer AI is missing.
Hey Product Hunt!
I'm Sourav, builder from India. Spent 6 years building AI products for others. Memoair is my first startup.
The problem I kept hitting: AI has no memory. Every session resets. You re-explain the goal, paste in the context, burn tokens just to get back to where you were. The most useful systems I built weren't the smartest ones. They were the ones that remembered.
So I'm building Memoair. A memory and context layer for AI that keeps context alive across sessions, so you stop starting from zero every time.
No funding. No growth team. Just us builders, the product, and a lot of figuring it out in public.
Closed beta is live and the waitlist is open. If you've hit this wall building with AI, I'd genuinely love to hear your experience. And if Memoair sounds useful, even better.
Replies
This is interesting. Feels like memory is less of a feature and more of a UX expectation now. If users have to keep reminding the system, it breaks the illusion pretty fast. Curious… are you seeing people want full control over memory or just want it to “just work” in the background?
@kabirsalunkhe Exactly, when the system forgets, the product stops feeling like a partner and starts feeling like a high maintenance tool.
To your question, in our discussions, we have seen builders need some control. If you cannot manage the memory, you cannot guarantee the output. That is why we are building a layer that handles the heavy lifting in the background while giving developers the "edit" option they need to keep things on track.
Have you run into this context reset problem yourself?
@sourav_das29 That makes sense…feels like it’s less about control vs automation and more about when control shows up. If users have to think about memory too early, it becomes work but if they can step in only when something feels off, it starts to feel like a real partner.
We’ve been running into this a lot while building CambrianEdge especially in multi-step workflows where context drifting is hard to notice until it’s too late. Still figuring out where that balance sits.
How are you thinking about surfacing memory to users without overwhelming them?
@kabirsalunkhe That is exactly the problem. When the AI starts losing the plot, it is usually because it got a small detail wrong five steps ago and kept running with it.
We try to keep the memory hidden so it does not feel like a chore. But we give the user a quick way to see what the AI currently thinks is true.
If they see a mistake, they can fix it right there before it ruins the rest of the work. It is like a map you only pull out when you feel like you are going the wrong way.
With CambrianEdge, does it usually get complex because it has too much information to handle, or just because the tasks have too many steps?
The 'most useful systems weren't the smartest ones, they were the ones that remembered' is such a good insight. im always reminding my ai what we spoke about
@olga_kargopolova That is the exact friction we are trying to kill. When you have to spend the first three prompts reminding the AI who you are and what you are building, it stops being a collaborator and starts feeling like a high-maintenance tool.
The goal is to move past that "Groundhog Day" loop where every session starts from zero. If the system actually carries the context forward, you can spend your time building instead of repeating yourself.