the protocol thesis: why self.md is a file format, not an app

by Ray Svitla

there’s a reason email won and every “email killer” lost. there’s a reason the web won and every walled garden eventually opened up or died. and there’s a lesson in both for anyone building tools for personal identity.

the lesson is: protocols beat platforms. always. eventually. sometimes it takes decades, but protocols win.

self.md is a bet on that lesson.

the protocol pattern

SMTP didn’t build an email app. it defined how email works. any client can send. any server can receive. the format is open. the network is permissionless. no one owns email.

HTTP didn’t build a browser. it defined how the web works. any server can host. any client can browse. the format is open. the network is permissionless. no one owns the web.

the pattern: a protocol defines the shape of communication. applications build on top. the protocol layer is free, open, and universal. the application layer is where competition, innovation, and monetization happen.

platforms try to skip the protocol layer. they say: we’ll handle the format, the network, the storage, and the application — all in one place. it’s easier, it’s faster, it’s more polished. and it works beautifully until you want to leave, or until the platform decides to change the terms, or until the platform dies.

your Notion workspace is a platform. your OpenAI custom instructions are a platform. your Claude memory is a platform. each one stores your identity data in a format you don’t control, on a server you don’t own, with export options that vary from “adequate” to “theoretical.”

the Noosphere warning

Gordon Brander tried to build a protocol for thought. Noosphere — a decentralized, self-certifying protocol for linked notes. the vision was beautiful: user-owned data, cryptographic identity, no platform lock-in. notes as sovereign entities in a peer-to-peer network.

Subconscious, the company behind Noosphere, wound down in 2024. the protocol never gained critical mass.

the lesson isn’t that protocols for thought can’t work. the lesson is that a protocol without a wedge is an academic exercise. Noosphere was technically sophisticated and practically inaccessible. you needed to understand self-certifying protocols and content-addressed storage to get started. the on-ramp required a PhD’s worth of context.

self.md learns from this failure. the protocol is a markdown file. you create it in any text editor. you don’t need to understand cryptography or decentralized networks or content addressing. you write a text file. that’s the wedge.

the sophistication comes later — routing, catalog matching, AI-powered analysis. but the entry point is a file you already know how to create.

why a file, not a feature

every major AI platform will eventually build “personal identity” features. OpenAI has memory. Anthropic has preferences. Google has… whatever Google is doing with your data.

these features will be good. they’ll be convenient. they’ll know things about you that are genuinely useful. and they’ll all be platform-locked.

your OpenAI memory doesn’t work with Claude. your Claude preferences don’t work with Gemini. each platform builds its own silo of your identity data, and switching means starting over. this is the textbook definition of vendor lock-in.

a file is different. a markdown file works everywhere. every AI model can read it. every editor can write it. every version control system can track it. the file belongs to you in a way that a platform feature never can, because the file lives on your filesystem and the feature lives on their server.

this matters philosophically — Illich would call it the difference between a convivial tool and an industrial one. but it also matters practically: the file is portable because markdown is portable. today it routes through Claude. tomorrow it routes through whatever model is best. the day after, it routes through a local model running on your laptop. the file doesn’t care.

the business paradox

here’s the part that makes investors nervous: how do you monetize a file format?

the answer is: you don’t. you monetize what sits on top.

the file is free. always. the self.md specification is open. anyone can create one. anyone can read one. anyone can build tools that consume one. the file is the protocol layer and protocol layers are free.

the value is in the intelligence layer:

the catalog — curated philosophies, approaches, tools, techniques, each with applicability conditions and contraindications. like a pharmacopeia for the mind. building a good catalog takes expertise, curation, and ongoing maintenance. that’s monetizable.

the routing — AI-powered matching between your identity state and the catalog entries that serve you. not simple keyword matching. nuanced, contextual, tension-aware routing that requires sophisticated inference. that’s monetizable.

the teams layer — understanding how different identity protocols interact. compatibility mapping. collaborative routing. organizational self-knowledge. that’s monetizable.

the protocol is the wedge. the intelligence is the product. the file is free because free files get adopted, and adopted files become standards, and standards create ecosystems, and ecosystems create value.

SMTP is free. Gmail makes billions. HTTP is free. Shopify makes billions. the pattern holds.

the never-lock-the-file rule

there’s one rule that overrides every business consideration: never lock the file.

the moment self.md requires our platform to function, we’ve failed. the moment the file format includes features that only our tools can read, we’ve failed. the moment a user can’t take their self.md to a competitor and have it work, we’ve failed.

this is the architectural equivalent of the burn test . can the file survive without us? if yes, the protocol is healthy. if no, we’ve crossed Illich’s watershed — from serving users to capturing them.

is this good business strategy? honestly, I don’t know. it’s good protocol strategy. it means that our advantage has to come from being better at routing, not from controlling the file. it means we compete on intelligence, not on lock-in.

every instinct in startup culture says this is naive. every instinct in protocol history says it’s the only approach that works long-term.

what a standard could look like

self.md today is a loose convention. a markdown file with some expected sections. that’s enough to start. but eventually, if adoption grows, it might need more structure.

not much more. the beauty of markdown is its informality. but maybe:

## identity
→ required. who you are. tensions, values, patterns.

## context
→ optional. current state, energy, mode.

## routes
→ optional. explicit routing rules.

## meta
→ optional. version, last-updated, privacy level.

a minimal schema. readable by any LLM. writable by any human. versionable by any git tool. no JSON schemas, no XML namespaces, no protocol buffers. markdown headers and natural language.

the standard should be so simple that someone could create a valid self.md without reading any documentation. just write about yourself in markdown. the structure is emergent, not enforced.

the bet

the protocol thesis is a bet. a bet that open wins. that portable wins. that user-owned wins. that the value of a universal identity file — one that works across every platform, every model, every tool — is greater than the value of any single platform’s identity features.

it’s a bet that the personal AI space will follow the same pattern as email and the web: platform lock-in works in the short term. protocols win in the long term.

I might be wrong. the major AI platforms might build identity features so good, so convenient, so deeply integrated that nobody bothers with a text file. the convenience might be worth the lock-in.

but I don’t think so. because identity is the one thing you shouldn’t rent. your preferences, your values, your tensions, your self-knowledge — these should live in a file you own, not a feature someone else controls.

that’s the thesis. a markdown file. an open protocol. and a bet that people will choose sovereignty over convenience when it comes to who they are.

→ identity as protocol — the philosophical foundation → convivial AI — Illich’s framework for tools that enhance autonomy → the three tests — the burn test as business constraint → why your second brain is dead — the problem the protocol solves

Ray Svitla stay evolving