Persona memory
A distilled sense of who you are and how you work — not a transcript, a model of the person it is helping.
AGPL-3.0 · your AI teammateA member that remembers
Persona memory — a distilled sense of who you are — and a hybrid vector + lexical graph of your code, grounded in one conversation that draws on both. AGPL-3.0 open core; it binds to localhost and holds one person’s memory of their own machine.
Get started$ self-host on localhost
AGPL-3.0
0
memories it keeps: a distilled persona and a graph of your code
1
conversation grounded in both — not a chatbot with amnesia
0
of it leaves localhost — one person’s memory of one machine
AGPL
open core; the hosted control plane is where entitlements live
What it is
One library, whole — not a platform you rent
A graph of your code
A hybrid vector + lexical graph of the codebase on this machine, so it reasons about your project, not the internet’s.
Grounded conversation
Every reply is grounded in the persona and the code graph at once — memory and context, not one or the other.
Binds to localhost
It runs on your machine and holds your memory there. Nothing is shipped off to be someone else’s training set.
Open at the core
AGPL-3.0. The member itself is yours to run and to read; the commercial control plane is a separate concern.
A control plane, when hosted
ai-member-cloud adds authentication, tenancy and entitlements over the AGPL core — the parts the localhost member deliberately lacks.
By the shape of it
A teammate who forgets you between conversations is a search box. Memory is the whole difference.
2 + 1
40% Persona
40% Code graph
20% Conversation
Two memories, one grounded conversation — persona, code graph, chat.
How it works
Three moves, in order
01
Index
Point it at your code. It builds a hybrid vector + lexical graph of what is actually on this machine.
02
Distill
Over time it forms a persona — a compact model of who you are and how you like to work.
03
Ground
Ask it anything. The reply draws on both the persona and the code graph, and never leaves localhost.
ai-member.ts
import { remember, ground } from 'ai-member';
// a hybrid vector + lexical graph of the code on THIS machine
await remember.index({ root: '~/code/app' });
// a reply grounded in the persona and the code graph — never leaves localhost
const reply = await ground(prompt, { persona, graph });“
A teammate who forgets you between conversations is a search box. Memory is the whole difference.