Nessie Wants to Give Your AI Conversations a Permanent Memory

The Macro: We Are All Generating Knowledge and Immediately Losing It

I had a conversation with Claude three weeks ago about pricing strategy for a side project. It was a good conversation. I worked through unit economics, tested different models, and landed on a freemium approach with specific tier breakdowns. That conversation is gone. I could not find it if I tried. It exists somewhere in a chat history I will never scroll back to, buried under dozens of subsequent conversations about completely unrelated topics.

This is happening to millions of people every day. ChatGPT has over 200 million weekly active users. Claude, Gemini, Perplexity, and a dozen other AI tools add tens of millions more. Every one of those users is generating original thinking in conversation with AI, and almost none of it gets captured, organized, or reused. The conversations vanish into infinite scroll. The insights evaporate. The work gets redone from scratch the next time the same question comes up.

The knowledge management market has been trying to solve the “capture and organize” problem for decades. Evernote was supposed to be your second brain. Notion was supposed to replace it. Obsidian, Roam Research, Logseq, and a constellation of tools built around linked notes and knowledge graphs have all taken a run at making personal knowledge management work. The pattern is consistent: people sign up with enthusiasm, spend a week building a system, then slowly stop using it because the maintenance overhead exceeds the retrieval value. The tool becomes another inbox to feel guilty about.

AI conversations are different from traditional notes in a critical way. They are already structured. When you have a back-and-forth with an AI about a technical problem, the conversation naturally contains the question, the exploration, the alternatives considered, and the conclusion reached. That structure is implicit in the dialogue. If you could extract it automatically, you would have notes that are better organized than anything most people write by hand.

The timing is interesting. We are at the point where AI power users have accumulated hundreds or thousands of conversations across multiple platforms. That is a corpus of personal knowledge that has real value but is completely inaccessible. Search within ChatGPT is basic. Claude’s conversation history is harder to navigate than my email from 2014. Cross-platform search does not exist. If you used ChatGPT for one project and Claude for another, those two bodies of knowledge are in separate silos with no bridge between them.

The Micro: Two Yale Grads Who Shipped an Android App at 13

Nessie imports your AI chat histories from ChatGPT, Claude, and other sources, then auto-organizes them into structured, searchable notes. You can search across your entire conversation history semantically, synthesize connections between ideas from different conversations, and share curated “AI brains” with other people. The pitch is Perplexity for your mind: a tool that makes everything you have already thought about findable and composable.

Anna Zhang is the co-founder and CEO. Yale BS class of 2024, former Amazon engineer working on recommendation services, with a research background in neuroscience at Rockefeller University. Tiger Wang is the co-founder and CTO. Also Yale BS 2024, also a former Amazon engineer, working on authentication systems. He shipped an Android app that hit 70,000 users when he was 13 years old. They are a two-person team out of San Francisco, part of Y Combinator’s Fall 2025 batch.

The early traction is promising. Within three weeks of launch, Nessie had over 1,200 users who imported more than 300,000 AI conversations. That is roughly 250 conversations per user, which tells me they are attracting heavy AI users, not casual ones. The 30% week-over-week organic growth rate suggests strong word-of-mouth, which makes sense for a product in this category. If someone imports their AI history and finds a conversation they forgot about, they are going to tell their friends.

The shareable brains feature is the one I keep thinking about. Imagine a tax accountant who has spent 200 hours working through complex tax scenarios with AI. That accumulated knowledge, organized and searchable, has value to other accountants. A developer who has debugged hundreds of issues across a specific tech stack has built a troubleshooting corpus that junior developers would pay for. The sharing mechanism turns individual AI conversations from throwaway interactions into durable, transferable knowledge assets.

The competitive landscape is sparse because the category barely exists yet. Recall.ai focuses on meeting recordings, not AI conversations. Mem and Reflect are AI-powered note apps but they require manual input. Notion AI adds AI features to an existing notes product. Nobody else is specifically going after the “your AI conversations are valuable, let us organize them” angle. The closest competition is the AI platforms themselves, if ChatGPT or Claude dramatically improved their built-in search and organization features. That is a real risk but not an immediate one, because platform companies tend to be slow at adding features that do not directly drive usage of their core product.

I have one concern. The value proposition depends on people having enough AI conversations to make organization worthwhile. That is a growing demographic but it is still a subset of all AI users. Many people use ChatGPT once a week for a quick question and never think about it again. Nessie’s market is the power user segment, the people running entire workflows through AI. That segment is growing fast, but the total addressable market today is smaller than the headline AI user numbers suggest.

The Verdict

I think Nessie is building for a problem that is going to get much worse before anyone else notices it. The volume of AI-generated personal knowledge is exploding and the tools to manage it do not exist yet. The auto-organization approach is the right call because manual knowledge management has a 20-year track record of failing for most people.

At 30 days, I want to see the import-to-active-use conversion rate. Getting people to import their history is step one. Getting them to come back and search it regularly is the real product. At 60 days, the question is whether the shareable brains feature gets traction. If people start creating and distributing specialized AI knowledge bases, that is a network effect that could drive growth far beyond the initial power user audience. At 90 days, I would want to know whether Nessie has built integrations with the major AI platforms or whether they are relying on manual exports. The easier the import, the lower the friction, and in a product like this, friction is the only thing standing between a useful tool and an abandoned signup.

The 300,000 imported conversations in three weeks tell me the demand is real. People want this. They just did not know they wanted it until someone built it.