start in one command

there is one command. it does the whole setup, and it tells you what it did as it goes.

npx innernet

paste it into a terminal (on a mac, press ⌘ space and type terminal) and press enter. you need node 18 or newer; node -v tells you. nothing is installed: npx fetches innernet, runs it, and forgets it.

what you'll see, step by step

before it does anything, it says what's about to happen: two steps anywhere, three inside a code project.

1. 1sign in. your browser opens innernet.live. sign in, or make an account; a new account meets its netti first and then comes straight back. the terminal waits for you, and prints the link as well, in case no browser opened (over ssh, say). when it's done you'll see signed in as and your email.
2. 2your ai tools. it looks for every ai tool on this machine that it knows: claude code, cursor, windsurf, codex, gemini cli, claude desktop, vs code, cline, roo code, opencode, zed, continue and antigravity. into each one's own settings it writes one entry pointed at your memory, and each line names the file. nothing else in those files changes; a server you already had stays where it is. claude and chatgpt live in a browser, so nothing on your computer can set them up. the run hands you the link and the few clicks instead; connect your ai tools walks through both.
3. 3this folder. inside a git repo it offers to start a memory for the project. say yes, and netti reads the readme and the file names (nothing else leaves your machine) and designs the pages that fit this project. then it adds two quiet git hooks, so every commit and pull is remembered, and a short note in AGENTS.md so coding agents know where the memory is. your repo in innernet has the whole of it.

then it tells you what now

the end of the run is the part to read.

• restart the tools that were open. tools read their settings when they start.
• five things to say, in plain words, in any of them. the first, load my projects from innernet, is the check that it worked.
• what to bring in. if claude code has been keeping its own memory on this machine, the run says so, with the one command that brings it over.
• where next: what to say to your ai, every guide, your dashboard, and npx innernet guide, the same tour in your terminal.

what it wrote

the key never leaves this machine except to talk to innernet, and you never paste it into a chat. if an ai ever asks you for one to "connect innernet", don't; give it the link instead and let the browser sign-in run.

commit .innernet/ and AGENTS.md (the run reminds you). that's how the link travels with the code: someone you've shared the memory with clones the repo, runs the same command, and reads the same memory.

running it again

running it again is always safe. it re-checks everything, fixes anything that drifted, picks up a tool you've installed since, and says all set when nothing needed changing. to see what it would do first:

npx innernet --dry-run

it shows every file it would touch and writes nothing, not even a sign-in.

the rest of the command

installed for good (npm i -g innernet), each is just innernet ….

when something looks wrong

• a tool isn't listed. it isn't where the command looks, or innernet doesn't know it yet. give it the link by hand: connect an ai.
• a tool shows ⚠. its settings couldn't be changed safely (a file that isn't valid json, say). the line names the file and what to add; fix it and run again, or add the entry by hand.
• the tool doesn't mention innernet. restart it, then start a new chat.
• commits don't seem to arrive. run npx innernet status inside the repo. it shows the last commit that reached your memory and, when one didn't, why.

every flag, in technical detail: the npx command.

next: what to say to your ai.