egoist vs innernet

egoist guards the door to your memory. innernet keeps what's behind the door current.

of everything we compare ourselves with, egoist's ai passport sounds the most like us. your context, owned by you, carried to the ai apps you use over MCP. their docs call it "a person's portable memory with an owner-controlled trust loop." the trust loop is the part they built with the most care, and it is very good.

the difference is what happens between the reads. egoist decides who may see your memory, for what and for how long. innernet's work is keeping that memory true as things change: what was decided, what replaced it and when, and who said each thing.

a consent and identity layer for personal context, made by Egoist Machines.

memory comes in two ways. an ai saves something and it lands as a proposal in your private inbox, and no app can read it until you approve it. connectors read live sources like calendars, notes and fitness trackers, and connecting one gives no app access by itself.

memory goes out through passes. an app asks for exact fields with a purpose and a duration attached, you see the company, the fields, the purpose and the duration before anything is shared, and a pass covers one app, one category, one purpose and one duration. passes don't widen on their own, you can revoke one at any time, and every read and proposed write is visible to you, kept as a record of what happened and when, never the memory itself.

approving a memory or a pass needs your own signed-in browser session, which a connected ai never holds. an ai can't approve its own memories.

it reaches Claude through the connector directory, ChatGPT as a published app, Gemini as a custom app, and any other MCP client by url. agents running on your own machine pair with a device code. there is an iPhone app, and apps can offer "log in with ai passport" on standard OpenID Connect. an optional memory lock seals your stores with keys derived from your passkey, one per category, and deleting your data shreds the keys.

one memory under the ai tools you already use, kept current by netti.

npx innernet connects the tools on your machine (Claude Code, Cursor, Codex, Windsurf, Gemini CLI, VS Code and more) to the same memory, and Claude, ChatGPT and any MCP client reach it at one url. what your tools capture, netti folds into maps: one for each project, and one for you.

egoist's public docs are careful about how memory gets in and how it gets out. we couldn't find how it is kept current once it is in: what happens when an answer changes, whether the old one is retired or kept, how long a fact stays true. if they do this, it isn't in the docs we read, and we'll correct this page when we learn otherwise.

that upkeep is most of what innernet does. a page describes the present. when something replaces it, the old reading leaves one dated line under that page's history, and the commits hold the rest. each kind of memory ages on its own clock: a law never, a decision slowly, a status in about two weeks. age decides what a tool reads first, and it never hides anything.

egoist's own example is a shopping app asking for your style, your sizes and your budget, for one purpose, for one session. that is a profile of you, read by apps that want to serve you better.

innernet holds the work too: each project's decisions, reasons, open questions and branches, read beside what it knows about you. one ask returns the parts of both that bear on it.

egoist asks per read: this app, these fields, this purpose, this long. innernet marks every personal fact with how far it may travel (private, self, trusted, work, public), and a connected app reads only trusted, work and public. your self and private facts stay in innernet.

ours is safe without asking. theirs is finer, and you can watch it work.

• ·consent you can see. a pass names the app, what it reads, why and for how long. that is finer than anything innernet asks today.
• ·a record of every read. every read and proposed write shows up, without the memory itself in the record.
• ·no ai approves its own memory. approval needs your signed-in browser session. in innernet a connected ai can write, and you see and undo it afterwards.
• ·reach. Gemini setup steps, an iPhone app, device codes for agents on your own machine, and a log-in button other apps can use. we don't have a phone app.
• ·a lock with your key. turn on memory lock and your stores are sealed with keys derived from your passkey, one per category. innernet's keys are held by innernet.

• ·the memory keeps itself current. netti folds what comes in, replaces what changed and dates it, and old readings stay in history instead of competing with the new one.
• ·memory ages by what it is. a decision stands until something replaces it, a status fades within weeks, and nothing is hidden for being old.
• ·your projects and you, together. decisions, reasons and branches beside the person who made them.
• ·who said it, on every fact. every personal fact records whether you said it, an ai reported it or netti inferred it, with the evidence it came from. your exact words are kept as yours only when they are yours.
• ·sealed at rest, by default. pages and facts are encrypted without a setting to find. the keys are ours to hold, so this guards against a stolen database, not against innernet itself.
• ·one press to leave. settings, data, export: after your password or an emailed code, everything comes out as markdown and json, and it goes back in through import.

use egoist if you want the apps you use, from assistants to shops, to know your preferences, and you want a say in every read: which app, which fields, why and for how long. or if you want it on your phone.

use innernet if your work runs across several ai tools and the hard part is keeping one true account of it: what was decided, what changed and when, and who said each thing.

both is fine. both speak MCP, and a tool can have both connected.

"egoist asks me before any app reads anything. why doesn't innernet?"

because today innernet decides by the fact, not by the read. every personal fact carries a class, a connected app reads only the trusted, work and public ones, and your self and private facts stay in innernet. that keeps the default safe without a prompt on every read, but it is coarser than a pass that names the app, the purpose and the duration, and it doesn't hand you a record of every read. egoist is ahead of us there, and we'd rather say so than pretend otherwise. what we add is the other half: a memory that stays current, dated, and clear about who said what.

accurate as of 15 october 2026, based on egoist's public site and docs at ego.ist. if we've described their product unfairly, tell us and we'll correct it.