Sociofabric

Philosophy

Life in this universe thrives on one idea: community. From herds of zebra grazing the savannah to pods of dolphins traversing the open ocean, the more intelligent life becomes, the more it needs the comfort of others. Community gives us purpose, a sense of safety, and permission to be vulnerable when we need it most.

For most of our history, community was handed to us by geography: the only one we knew was the one we were born into. Only very recently has moving become ordinary, and as we learned to live among strangers, we discovered how hard it is to keep the thing that matters most. We learned how easily we fall out of touch with people we once knew by heart. We learned how easy it is to feel alone in a crowded foreign city.

Maybe social media was meant to solve this. For a good while, it did. It offered connection at a scale nothing else could. But over time, these platforms began optimizing for advertising, and a conflict of interest settled into their core: monetized attention versus free connection. Because the freemium feed was all we knew, we accepted the terms. If we wanted to stay close to the people we love, that was the price. And the feed drifted ever further from what we came for: celebrities and influencers performing lives so removed from our own that we felt inadequate; friends and family buried beneath posts from people we had never met. We ended up sitting around a television we didn't program, except this one was also the only way we stayed in touch.

I wanted to change that. I believe the sense of community is inseparable from the sense of self, and a human need this fundamental should never be gated behind entertainment and advertisements. Everyone deserves a dignified life, and to me, a dignified life comes with privacy, community, and belonging.

So I built Pixie: an app where you own your data, you own your on-device AI, and you own your connections, arranged the way they actually work for you. Pixie will never show ads, and it will never compromise on your privacy.

And there has never been a better time to build this. On-device models that can run on an iPhone, Apple's own Foundation Models among them, finally support image processing and tool calling starting in iOS 27: elements essential for parsing documents and taking action. Compute is scaling sufficiently for me to believe that within a year or two, even bigger models could run on an iPhone at 20+ tokens/second. Today, you can use BigBrain to leverage the bigger models running on your desktop for compute on the go (and this paradigm will likely continue as desktops themselves run bigger and better open source models). The time is ripe for the cryptographic guarantees too: the post-quantum primitives Pixie is built on were standardized only very recently, so a network built today can outlive a post-quantum reality. The skill ecosystem is alive and well in developer communities, with open source skill repositories like superpowers gaining significant popularity. Pixie opens up this skill customization community to the broader user base, allowing for building and sharing skills that the community finds useful.

I also believe Pixie is the best assistant you can carry today, not because it runs the biggest model (it may not), but because it does things few assistants can. Installed across your devices, Pixie uses InstaFetch to find a file on your computer and hand it to your phone, no cloud storage involved, and BigBrain to put the stronger model on your desk behind the questions you ask on the go. It tends your relationship graph quietly, and speaks up when someone you care about is drifting out of reach. Huddle keeps you posted on the people closest to you, talking agent to agent with theirs. Puppeteer even lets you point an external system like Claude Desktop at Pixie and hand it the controls: reading and sending messages, watching over your relationship data, working in your notebook, always with your consent. You can build your own skills from the tools Pixie exposes, run any MLX community model you download from Hugging Face, rewrite the system prompt until Pixie behaves like your Pixie, and share all of it with your friends, protected end to end by post-quantum encryption.

Join me in this journey, as we explore this new frontier: AI-augmented, but deeply human, relationships.

What we built

Promises the architecture keeps

Yours, permanently

Your companion, memories, and relationships belong to you. Cancel any time — nothing is deleted, and one tap packs it all into an encrypted archive only you can open.

Ad-free

Your attention was never the product. It never will be — no profile of you exists for anyone to sell.

On-device

The model runs on your phone. Your memory lives on your hardware, moving between your devices only under seal.

Post-quantum encrypted

Messaging and contact discovery are quantum-safe from the first byte. Recorded today, still unreadable tomorrow.

Consent-first

Pixie notices, remembers, nudges — then hands you the phone. Nothing reaches another person without your yes.

Model-agnostic

Run any model locally, or reach for a bigger one — your choice, never a default. We never own the mind behind your companion.

Verifiable

The protocol is published, the proofs machine-checked, the wire format specified down to the byte. What you can't yet verify — that the app in your hand was built from this code — is written there too, as work owed. Don't trust us — check.

Check it on GitHub