Artificial Personalized Intelligence

does your AI give you the same answer it gives everyone else?

that's what a GENERAL intelligence is built to do. it serves everyone, but no 2 humans are the same.

your AI is an extension of you, your skills. the more it knows how you think, the better it serves YOU. so aim for A-P-I: Artificial Personalized Intelligence.

what makes an AI yours

1. a profile it reads first. who you are, what you do, how you talk. without it, every chat starts from almost zero on you, or from a hazy, irrelevant memory your AI stored over time.
2. your edits. every correction you make is a suggestion or rule it missed. have your agent log what you change and the things it gets wrong constantly, and have it learn from them.
3. how firmly you meant it. "i like" is a preference. "never" is a rule. spell out the weight behind your words so 1 offhand comment doesn't become a law your AI applies everywhere because it thinks you meant it.
4. memory you can open. if you can't read it and delete a line, you don't control what it thinks it knows about you. most LLMs now have a memory log in settings. review it, and have your AI edit or delete what's wrong.

how i do it

this is the worked example, from Auny's own setup.

her AIs pull new and updated context about her from her vault whenever they work on a task: who she is, what she does and how she writes. plus a behavior doc that tells them how to act with her. all of these are graph folders that update automatically from what she does, says and works on with Claude.

her Claude logs the way she talks to it, verbatim, in a ledger. so it knows she says "algo" not "algorithm", that she curses a lot when she's angry, when to push back, and when not to test her and just get the job done.

it also weighs how firmly she said something: a gesture, a preference, a practice or a rule. when she said her emoji count is "circumstantial", it stopped flagging posts with 5 emojis as "you used too many emojis. was this deliberate?" while running analysis on her content.

it has ~500 memory files from working with her. every one is a file she can read, review, and tell it to update or delete if it's wrong.

and when she corrects the same mistake more than twice, it becomes a guard: Claude double checks itself on that before any output.

[it sounds like a robust system. it started because re-explaining yourself every session is a waste of your time]

a real one from my workflow

her Claude prepped a first draft of her interview article with @suavecito585. she edited it and rewrote it her own way.

then she pasted her version back and told it to take notes on every correction she made.

it came back with 4 new rules it could record, and 3 it already had that the rewrite confirmed again. some of them:

all of it went into her article notes, which her Claude reads before it drafts the next article.

try it on your next AI draft:

1. edit it until it sounds like you
2. paste your version back and say: "list every change i made and the rule behind each one"
3. save those rules where your AI reads them first: custom instructions, a CLAUDE.md or a project file

start yours in 1 command

the personalization protocol is open source: agent-personalizer on Github

npx agent-personalizer

need to change something later? edit LEARNED.md (or tell your AI), then run it again. it updates its own files and keeps your edits.

your homework

ask your AI to describe you in 3 lines. if that description could fit 1000 people, it's not personalized yet 👀

then answer the questions in the installer and ask it again.