Epicenter Thinks Your AI Should Remember You, and You Should Own the Memories

The Macro: Your Digital Life Is Shattered Across 40 Apps and None of Them Remember

I want to describe a problem that is so pervasive most people have stopped noticing it.

You take meeting notes in Notion. You record transcripts in Otter. You manage tasks in Linear. You chat in Slack. You write code in VS Code. You email in Gmail. You talk to Claude or ChatGPT for help with all of the above. Every single one of these tools knows something about you, your work, and your context. And none of them share that knowledge with each other.

The result is that you spend a staggering amount of your day re-explaining things. You tell your AI assistant the same background about your project that you already wrote in your project doc. You re-describe your preferences that are already evident from your chat history. You manually copy context from one tool to another, acting as a human API between apps that should be talking to each other.

ChatGPT tried to solve this with its memory feature, and it was a good instinct. But the implementation locks your memory into OpenAI’s servers, tied to a single product, controlled by a single company. If you switch to Claude or Gemini or whatever comes next, your memory does not come with you. You start from zero.

The local-first software movement has been building toward a different answer. Tools like Obsidian, Logseq, and Standard Notes have demonstrated that people want to own their data, not rent access to it. The missing piece has been a shared layer, something that lets multiple local-first apps access the same pool of context without routing everything through a cloud service.

This is where the opportunity gets interesting. Rewind (now Limitless) went after personal context capture but built it as a cloud-dependent, proprietary system. Apple Intelligence keeps everything on-device but only shares context within Apple’s ecosystem. The open, portable, local-first version of this idea, one that works across any tool and any platform, does not really exist yet.

The Micro: A Solo Founder With 10K Commits Per Year and a Very Clear Philosophy

Epicenter was founded by Braden Wong, a Yale graduate who averages roughly 10,000 commits per year on open-source projects and previously worked at three Y Combinator startups. He came through Y Combinator’s Summer 2025 batch as a solo founder based in San Francisco.

That commit volume is not a vanity metric. It tells you this is someone who builds constantly, publicly, and in the open. The entire Epicenter philosophy flows from that ethos.

The product is an ecosystem of open-source, local-first apps that share a memory. The storage layer is plain text files and SQLite databases that live on your machine. You own them. You can read them with grep. You can version them with Git. You can open them in Obsidian or VS Code or whatever tool you prefer. There is no proprietary format, no cloud dependency, and no vendor lock-in.

The tagline is “A Database for Your Mind, Built on Plain Text,” and it captures the vision precisely. Your transcripts inform your notes. Your notes guide your AI. Your AI learns from your entire history. And all of it lives in a folder on your computer.

The first product built on this foundation is Whispering, an open-source speech-to-text application built with Svelte 5 and Tauri. It supports both local and cloud transcription models without requiring a subscription. The second is epicenter.sh, a codebase query interface that runs a local OpenCode instance with secure tunneling.

What makes this architecturally interesting is the compatibility story. Epicenter is not asking you to abandon your existing tools. It works alongside Obsidian, VS Code, Git, grep, Claude, and OpenCode. The pitch is not “use our app instead” but “use our memory layer underneath everything you already use.” That is a much easier adoption path than asking people to switch tools.

The Rust and Svelte 5 technology choices are deliberate. Rust gives you native performance and memory safety for the local-first components. Svelte 5 gives you reactive interfaces without the overhead of React. Both choices optimize for speed and resource efficiency on the user’s machine, which matters when you are building software that runs locally rather than in a data center.

There is a Discord community and active GitHub repos, which is the right engagement model for an open-source local-first product. The people who care about data ownership tend to be the same people who evaluate software by reading the source code.

The Verdict

I think Epicenter is betting on the right long-term trend. Data portability and local ownership are values that only become more important as AI gets more powerful and more personal. The people who build their digital lives on open, portable formats are going to have dramatically more leverage than those who are locked into any single vendor.

What I would watch at 30 days: adoption velocity. Local-first and open-source products have a discovery problem. The people who want this product the most are often the hardest to reach through traditional marketing channels.

At 60 days: the ecosystem effect. One app with shared memory is a nice demo. Five apps with shared memory is a platform. The value proposition scales non-linearly with the number of apps that can read and write to the shared context layer.

At 90 days: whether the solo founder model holds. Building an ecosystem is a massive undertaking, and 10,000 commits per year is impressive but there are only so many hours in a day. The open-source community could become force multipliers, but managing a contributor community is its own full-time job.

The plain-text-and-SQLite foundation is genuinely inspired. It is the kind of design decision that looks boring on paper and turns out to be the most important choice the company makes. I am rooting for this one.