News for hackers. Stuff that matters.

andrei_gusev

Re:Ask HN: What do you all use to share context with team mates? on Wednesday November 18, 2026 @09:38AM (#50546851)
I came to the shared memory problem from the problem of AI's bad memory in general. I was renewing my business visa in Japan, lots of documents and bureaucracy, and Claude kept forgetting numbers and mixing up dates. So I set up a server + vector DB + MCP in the cloud, with one tool for saving and one for reading. I saved all the documents, and in a new chat, even from my phone, Claude started answering correctly.

From this I grew a whole project, memocore.ai. Same idea: agents connect over MCP, and it's split into workspaces (you invite users there to work in the same context) and projects. I moved all my work there. At any point in a chat I write "save", and it finds the right project from the context and saves it. If the project has other members, they see it right away and can use it on their side. We save everything to this knowledge base: daily standups, tasks, business logic, decisions made in chat. One colleague does research, another picks up the research and writes a task, the next one builds the feature and saves the docs. And all this knowledge is connected.

Re:Google's AI answers reducing search traffic by 40% on Tuesday November 17, 2026 @09:55PM (#50541279)
Looking at the analytics of my product (memory for AI agents), more and more traffic is starting to come from AI assistants: ChatGPT cites us, suggests our plugin, and users come. Same with Claude. ChatGPT already runs ads in its answers. Haven’t tried it yet, but it looks interesting. Google has already put AI into search, and I think soon it will integrate deeper and deeper with Gemini. Looks like everything is heading toward doing content, services and purchases all through agents, without visiting other sites.
Re:Mem Is Shutting Down on Tuesday November 17, 2026 @09:22PM (#50541071)
Considering how much we talk to AI agents now, I think a separate notes app is no longer relevant. Ideally, the agent should remember everything. Though there are still a lot of problems here. I ran into this when I was renewing my Japanese business visa, and that thing is very bureaucratic. So I spun up a RAG in the cloud and connected it to my Claude over MCP. It made me happy, I got really hooked on it, then got my colleagues hooked, and in the end turned it into a full service.
Re:Thoughts on memory layers like mem0 on Tuesday November 17, 2026 @09:36AM (#50531953)
There is a market, but for now it's all concentrated in business and teams, especially in enterprise solutions. I'm building a similar service.

When I was renewing my Japanese business visa, and that's a very bureaucratic thing, Claude's memory really let me down. It kept forgetting different numbers, mixed up dates, which documents were done and which weren't yet. I built a small project for myself, and it helped a lot. Then I rolled it out for everyone.

It's growing organically, from Google, from the ChatGPT and Claude directories, so users come with a problem and solve it. Mostly these are teams and small business owners. But when I tell people about a separate memory layer, they say their Claude or GPT handles it fine. So there is a market, but not among individual users.