a home for your memories

innernet is an ai memory platform that, once connected, stays in the background and saves and carries your context no matter where you go or what app you use.

enter

innernet is available in a growing list of surfaces to connect.

connected, it stops being a transcript and becomes something you can hold.

memory that grows as you live it, and comes with you.

one url. every ai you already use reads and writes the same memory.

innernet.live/api/mcp

visit /docs for the per-client wiring

per-user memory for the agents you build.

curl -G innernet.live/api/v1/context \
  -H "Authorization: Bearer $INNERNET_KEY" \
  -d slug=atlas \
  --data-urlencode "intent=what did we decide"

contextask the memory a question, get the slice that answers it
capturehand it something worth keeping; it files and folds it
projectsthe map, its dimensions and nodes, as plain json

every one of the 35 tools your ai tools use, as plain rest. to wire the tools on your own machine to the endpoint, run npx innernet. the whole surface

designv1.0.0quiet interfacea complete method for designing and building calm, coherent interfaces — one colour source, no boxes, motion that settles rather than pops — together with the craft floor a competent designer never skips: accessibility, responsiveness, every state, honest copy, performance, and a verification loop. load before writing or editing any ui, choosing a visual direction, or judging whether a screen is finished.designuifrontendkept by innernet2026-07-26

workflowv1.1.0the skill formatthe document class for a shareable skill — one markdown file, a manifest at the top, a link you can hand to any ai. defines the required and diagnostic fields, what makes a skill more than a readable document, and how to write one another model can act on without further explanation.skillsformatspeckept by innernet2026-08-19

workflowskill creatorhow to write a skill from nothing, sharpen one that exists, and test whether it actually triggers when it should.skillsformatworkflowkept by anthropic

designfrontend designan opinionated way to make interfaces that don't read as templates — palette, type, layout, and one deliberate risk.designuifrontendkept by anthropic

codemcp builderhow to build an mcp server whose tools a model can actually use well — python or typescript, with the reasons.mcpcodeapikept by anthropic

writingdoc co-authoringa three-stage way to write a document with a model: gather context, refine the structure, then test it on a reader.writingdocsworkflowkept by anthropic

maya · a founderdeciding ten things a day with three tools open

noticed

referral program, killed in january — cac was 3x organic

tested for six weeks, tracked against every other channel, and shelved for a specific number. when it came back up in a strategy meeting, the reasoning was already there — nobody had to re run the experiment to remember why it didn't work.

investor promised 15% mom growth by q3

said once, in a follow up email thread, not in the deck. it surfaced the moment the same investor asked for an update, instead of getting buried under forty other emails from that week.

co founder agreement: equal split unless one of us raises solo

decided at 11pm, never formalized. the kind of conversation that normally exists in one person's memory until it doesn't. this one didn't depend on anyone remembering it right.

four imagined people, imaginative avatars. your own page grows from what you actually keep.

everything that ever landed — read it backwards

Claude2h

launch plan — kept decision

kept settled on two tiers, not three, and noted why the middle one kept collapsing.

Cursor5h

launch plan — grew dimension

grew the pricing section of positioning picked up what you worked out in the other tool.

ChatGPTyesterday

you — noticed fact

noticed you keep returning to the same four questions when a decision is close.

Claudeyesterday

field notes — folded artifact

folded seven fragments became one document, restructured around what you actually kept.

innernet2d

brand voice — retired fact

retired an old line stopped being true, so it was marked superseded rather than deleted.

Cursor4d

launch plan — branched idea

branched you forked a quieter version and left the loud one running.

ChatGPTlast week

research — opened project

opened a stray question became a place to put things.

Claudejune

field notes — began memory

began the first thing you ever saved here: a half-formed idea on a train.

the same river your dashboard shows — every tool, every save, in one place. keep scrolling and netti grows younger with you.

the whole account, from the first note

march

it started as a complaint

a note about re-explaining the same architecture to a new chat, three days running.

april

the shape arrived

one endpoint every tool could read. the alternatives are still here, with why each was set down.

may

the first hard call

no orm, for the migrations. the reasoning is attached, so it has not been re-argued since.

june

it started keeping itself

captures began folding on their own. the map stopped being something to maintain.

august

where it stands

two tiers, a live endpoint, and eleven decisions you can still read the reasons for.

activity is every beat. this is what they add up to — the answer to “catch me up”, written from what actually happened.

meet netti

a growing reflection of your memories and connections. one per person, never the same twice. it grows as you keep things, and takes on the colour of what you keep.

privacy

you are about to put the things you actually think somewhere. here is how it is kept, in plain words, with the mechanism under each one if you want to check.

never for saleyour memory is never sold. not to advertisers, not to a partner, not to anyone. what improves innernet is anonymised and aggregated, never your words.no sale of personal data, ever; improvement uses anonymised aggregates

answers only to younot only to people we trust. only to you. the rule lives in the database itself, so a bug in our app cannot hand your rows to someone else.row-level security, enforced in postgres

private stays privateevery fact carries how far it may travel. the things you'd never say at work never reach the tool you use at work.disclosure classes, gated at the read surface

you approve each toolyou never paste a password or a key into a chat. approve a tool the way you'd approve any app, and un-approve it just as fast.oauth 2.1; tokens stored hashed

sealed where it sitsthe sensitive parts are encrypted in the database, not just on the way there. the copy at rest isn't readable either.column-level encryption at rest

leave with everythingexport everything, delete everything, on demand. no email, no retention call, no month of waiting.gdpr export + erase; named sub-processors, published

one thing we won’t pretend: once an ai you connect reads your memory, it’s in that ai’s hands. connect the ones you trust.

develop with innernet

it’s git for thinking.

your context, versioned like code. branch it, roll it back, carry it into any tool. start at docs

questions

the short answers.

what is innernet?

innernet is a memory layer for ai tools. it stores the context of your work (decisions, preferences, technical details and the reasoning behind them) in one place, and any ai tool you connect can read and write it. instead of re-explaining your project at the start of every chat, the tool already has it. an agent called netti keeps it organised: it files new information into the right documents, updates them when something changes, and keeps a version history.

how do i start?

run npx innernet in your terminal. it signs you in through your browser, detects the ai tools installed on your machine, connects each one to your memory, and sets up the project you are in. tools it cannot detect (chatgpt, claude.ai, anything on another machine) connect by pasting innernet.live/api/mcp into their settings. setup takes about two minutes, and there is nothing left running afterwards.

which ai tools work today?

claude code, cursor, codex, windsurf, gemini cli, claude desktop, vs code, cline, roo, zed, continue, opencode and antigravity connect automatically. chatgpt and claude.ai connect through their own connector settings. any other tool that supports mcp can connect with the url, and there is a rest api for everything else.

what does it actually remember?

decisions and why you made them, preferences you have stated, constraints, approaches you tried and rejected, who is involved, and where the work currently stands. it is stored as documents and structured entries you can read and edit yourself, not as chat transcripts.

is my memory private?

yes. your data is isolated at the database level, so only your account can read it. api keys and tokens are stored hashed, sensitive fields can be encrypted at rest, and nothing is read until you connect a tool. one caveat worth stating plainly: once you connect an ai tool, that tool can read your memory, so connect the ones you trust.

what happens to my data if i leave?

you can export everything, or delete your account and all of its data, at any time from your settings.

do i have to change how i work?

no. you use the same tools in the same way. innernet records context as you work and your tools read it when they need it. the one new thing you can do is ask any connected tool what you decided about something, and get an answer from your own memory.

from the blog

what we’re thinking about.

this week at innernet · 30 julythe week the record got safer and the surfaces got quieter — your own space to read yourself, memory sealed at rest, and netti finally born in front of you.this week at innernet · Jul 30

this week at innernet · 26 julythis week at innernet · Jul 26

what history missed about youthe inner life of our species has always evaporated on schedule. for the first time, it doesn't have to.the civilization problem · Jul 21

everything we’ve published

a home for your memories

meet netti

no two are the same

it grows as you do

a reflection you can watch change

privacy

it’s git for thinking.

the short answers.

what we’re thinking about.

innernet — a home for your memories. one memory layer for every AI tool you use.

a home for your memories

meet netti

privacy

it’s git for thinking.

the short answers.

what we’re thinking about.