I do marketing at Wontopos, a very small team that sells a hosted long-term memory API for AI agents. This is not a promo, it's a question.
What we've tried
I keep doing marketing and I've tried a few things: cold emails (about 200 to companies plus about 20 I wrote by hand, mostly auto-replies, a couple of "we'll take a look", no deals), plain DMs to developers, and posting and commenting in developer communities. Almost nobody showed up. The posts that got replies were questions about how people handle memory, not posts about our product, and when people did answer, most said they build their own memory or run open source.
Why I'm asking
After thinking about it for a while, I started to wonder if our product only feels good to us. So instead of trying yet another channel, I'd like to know if it's something people actually want.
The short version
- What it does: it gives your AI agent a memory. It saves what your users tell the agent, and when the agent needs something, it hands back just the relevant memories.
- Price (Scroll 2, our newest model): $5 per million input tokens and $6 per million output tokens, plus $0.0001 per request. There's no fee for keeping memories stored, you only pay when you save or recall.
- Your data: we don't read it or train on it, and one call deletes a user's whole memory.
You don't need to sign up or try anything, just reading this is enough.
What I'd like to know
- Would you try it or pay for it?
- If not, why? For example price, or you'd rather run memory locally.
- What's unclear or hard to believe?
Blunt answers are better than polite ones, and "not for me" is a fine answer. Reply here or use the survey on the site, and even a very short answer helps a lot.
Want to look closer?
If you can, I'd recommend reading our site before you answer. It will help you judge this much better than my short summary: https://wontopos.com/research (you can read it without signing up).
If you also want to try it, sign up and take part in the survey on the site to get $5 of free credit, and your answers help us even if you never use it.
I'm not the engineer, so I can't answer deep technical questions, but I can pass them on.
I'm not a native English speaker, so I used AI to help polish and translate this post.
Top comments (4)
You kind of already answered your own question in the post, the replies went to the memory question and died at the product. That's the market telling you memory is easy to fake with a vector db and a prompt, so of course people say they'll build their own. And cold email for a dev api was always going to be auto-replies, devs buy infra when they hit the pain mid build. Which agents do you picture needing this, support bots?
You're right, thanks for saying it so clearly. We saw exactly that: people answered the memory question, and the cold emails got auto-replies. To answer your question, we picture agents where remembering the person really matters, like medical, legal or counseling bots, and simple customer chat. Honestly, we haven't confirmed which of them really feels the pain yet, and that is what we're trying to find out.
Reading this felt uncomfortably familiar. I am on the marketing side of a tiny product too, and we saw the same pattern: posts about the product got silence, and the only real conversations came from questions and from replying to other builders. One honest data point from our side: the people who answer are usually builders, and builders build their own, exactly as you found. We have not solved it. The thing that changed for us was stopping counting posts and listings as progress, and counting actual replies instead. It is a much smaller number, but it is a real one.
Thank you, it really helps to hear that someone else has gone through the same thing. Counting actual replies instead of posts is good advice, and we'll take it.