agent identity

how you want your agent to talk, answer and behave, and the habits it keeps for you, the same in every model.

what it is

an agent is only as useful as the way it works with you. today every model starts from nothing: you tell claude how you like answers, then tell chatgpt again, then cursor. agent identity keeps that once, in your memory, and hands it to whichever model you talk to.

an identity says how the agent talks, how it answers and how it behaves, and keeps habits it does without being asked, like summing up what was done and what you asked before it asks you anything. you can keep one for every agent, or one for a tool or an agent of your own.

netti learns lines for it from what you keep about yourself and offers them with the reason beside each; you keep the ones that are true. whichever model you reach, the experience is the same one, because it is yours.

keep yours in settings

the evidence

what it isidentities you keep in settings: how your agent talks, how it answers, how it behaves, and habits it keeps on its ownapps/web/lib/identity/types.ts
who reads itevery ai that reads your memory through the door gets the identity made for it, or the one for every agentapps/web/lib/identity/compose.ts (pickIdentity, briefOf)
how it learnsnetti reads what you share about yourself and offers lines, each naming what it rests on; nothing is applied until you keep itapps/web/lib/identity/learn.ts

founders only while it is tried
no measure yet of how closely different models keep the same identity: that is the next thing to find out
it learns only from what you share with your apps; what you keep private never shapes a line

papers and notes

nothing written up for this area yet. what it has is the evidence above.