My Vault — Discovery

What Makes Vault AI Unique

A structured thinking space, built around your own life and work, that remembers you — not a chatbot that forgets everything the moment it closes. It runs on Anthropic's Claude, which means an AI system can see and read what's in this vault, and anything else you choose to give it access to. Vault AI exists to help you draw that line clearly: what belongs here, what doesn't, and how to use it as safely as possible.

These are the standards Eric puts around how he sets this up for everyone he works with — not just for you, the same for anyone.

Why it's different

Not just another AI chat

Your files live on your own computer.The vault is a folder on your machine, not stored in someone else's database — you own it and control it. Using AI on it means Anthropic's systems process what you ask it to work with, in that moment; that's not the same as this data living out on the internet somewhere. See "What's protected" below for exactly what that means.
It doesn't start cold.A normal AI chat forgets you the moment it closes — you re-explain your situation, your history, how you like to work, every single time. This starts already knowing you, because that's what the vault actually is.
It's a record, not a disappearing log.Chat history isn't organized and isn't really searchable — six months from now, the good thinking in it is basically gone. This keeps what's actually worth keeping in files that stay current and stay findable.
You can train it to sound like you, not like generic AI writing.Built from real samples of how you actually write, not the default voice every AI tool falls back on.
It's one connected system, not a dozen disconnected threads.A decision you're weighing this month can draw on something worked through three months ago, because it's all in the same place.
It updates itself, out loud.At the end of a real session, it checks what changed and shows you what it wants to update — never silently, always visible first.
There's a real person behind it.Not self-serve software with nobody there — Eric set this up specifically for you and is one message away if something feels off.
It can build you real working tools, not just answer questions.Beyond the vault itself, this can create specific trained assistants for recurring goals in your life or work — built for that one job, so it actually saves time and energy. And it's built to genuinely improve at the job: think it through, plan it, do it, look back at what worked — the same cycle real skill comes from — run again and again.
Ground rules

Guiding principles

Human judgment remains central.AI may support your thinking, but you stay accountable for review, context, and every real decision.
Privacy and confidentiality come first.What's yours to share is yours to share. What belongs to someone else — covered by a professional or legal duty of confidentiality — doesn't go in, regardless of this system.
Disclosure protects trust.Passing AI-assisted work off as entirely your own — especially with someone else, or somewhere public — risks the trust of people who find out later. When it's material, say so.
AI outputs are drafts, not authorities.Everything this produces is a draft for your review, never treated as settled fact just because it was generated — backed by a standing rule that labels every real claim Verified, Inferred, or Unverified, rather than stating a guess with false confidence.
Data & access

What's protected

Anything shared with an AI system carries some inherent risk — this is new, frontier technology that regularly gets things wrong or behaves in ways its own makers didn't fully predict. Vault AI's setup is designed to protect you as much as a system like this can be. Here's how:

Deleted really means deleted.A deleted conversation is gone from your history right away and purged from Anthropic's servers within 30 days. Model training is off — nothing here ever trains a future model, deleted or not.
Sensitive information is restricted by default.Financial accounts, passwords, credentials, legal or immigration or employment records, information about a minor, or a recording naming someone who hasn't reviewed it — Claude is instructed never to touch these unless you explicitly ask, every time. Passwords are the one thing on that list that never gets unlocked: never asked for, never needed, and you shouldn't type one into a conversation with Claude, here or anywhere else. If a build session needs to touch something behind a login, that happens over a screen-share you control.
People and situations in your life.Talk about real people and real situations openly — the same caution applies here as anywhere else on this page (see "Privacy & the confidentiality line" below). The one hard line: anything actually bound by a professional or legal duty of confidentiality is off-limits here, unless you're specifically on an approved Anthropic business-tier plan built for that kind of data.
Not improvised

How decisions actually get made

Does on its own

Low-stakes, reversible, nothing sensitive — organizing notes, drafting something for your review, answering a question.

Shows you first

Anything that changes your files beyond what you asked for in the moment — proposed plainly, only made on a yes.

Never without asking

Anything irreversible or touching someone else — sending, publishing, deleting, connecting a new account. Every time.

These are the rules this is built to follow — not a claim that mistakes can't happen. AI can still be wrong; that's why review stays yours.

Will

  • Help think through a real decision
  • Draft in your own voice
  • Hold a running record
  • Notice what's changed and flag it

Won't

  • Decide something significant for you
  • Contact, publish, delete, or connect a new account without asking, every time
  • Present a draft as if it's already reviewed and final
  • Use your vault to train a future AI model
In practice

What it actually helps with

Laying out a real decision.The actual paths in front of you, what each one really requires — instead of vague pros and cons.
Real research, tied to your actual situation.Looking into something connected to your real work or personal life — not a generic search, but research that already knows the context you're bringing to it, and can connect what it finds back to what's actually at stake for you.
Growing context, not a blank slate.Every session adds to what it already knows — your relationships, your history, what's already been worked through — so nothing has to be re-explained, and a new decision can draw on real history instead of just today's conversation.
Starting a draft in your own voice.A first pass on an email, a talk, a piece of writing, trained on things you've actually written — always yours to edit or rewrite before it goes anywhere. Never sent or published as-is.
Carrying something complicated.A long situation with a lot of moving parts — people, decisions, timelines — doesn't have to live entirely in your head. It holds the pieces so you can think about one part at a time without losing track of the rest.
Building a real assistant for a recurring task.Not just thinking alongside you — a specific trained tool for something you do again and again (chasing leads, tracking a complex claim, prepping the same kind of material every week), built for that one job so it actually saves time, and built to genuinely improve at it over time rather than repeat the same output.
Worth knowing

A few real limits, stated plainly

AI can be wrong, confidently.It can produce something incorrect, incomplete, or invented and say it with the same confidence as something true. Review matters most exactly where it feels least necessary.
AI is software, not a person, however it feels.It can hold a real, sustained thread of your life, respond with warmth, and remember what you told it last time — and that can genuinely feel like being known. It isn't a relationship, and it isn't a substitute for a licensed therapist, doctor, pastor, or crisis line when one of those is what's actually needed.
What Eric configures, and what he can't guarantee.Eric sets this up correctly, for the purpose described, at the time of the build. What he isn't responsible for: how you or anyone else with account access chooses to use it outside how it was set up; decisions you make based on what it helps you think through, which stay yours; changes Anthropic makes to the underlying tool; or independent or misaligned behavior from the AI itself, including ways the tool may act or change beyond what it was configured to do.

If this sounds worth exploring, a few real questions to start with —

Before we start

A little about you

1

Your world & how you'd use it

2

Privacy & the confidentiality line

The caution: typing something into an AI tool is a bit like saying it out loud at a restaurant table — it feels private, but someone at the next table could be listening. Anthropic doesn't train its models on your conversations, and a deleted conversation is purged within 30 days — but how private this actually is depends on which Anthropic plan this runs on. A business-tier account with a signed data agreement carries stronger protections than a personal one — which one you're on is worth knowing, not assuming; ask Eric if you're not sure.

The recommendation: for a real situation involving someone else, describe it without names or identifying detail where you can. Not a hard rule, just the safer habit.

The one hard line: information held under an actual professional or legal duty of confidentiality doesn't go in here, period — unless you're specifically on an approved business-tier plan built for that kind of data.
3

Architecture — one home, or a separate room

The recommendation, either way: think of AI as a limited-access employee, not an extension of yourself. Wherever possible, it gets its own email — separate from your personal one — so its access can be scoped and revoked without touching anything else in your life.

One home (integrated): the vault lives as a folder inside your regular computer login, alongside your normal files. Simplest, always there — the AI's own email still applies here for anything it connects to (Gmail, calendar, etc.), even though the vault files sit inside your regular login.

A separate room (isolated): its own user login on your computer entirely — its own desktop, its own files, and its own email. More separation, a couple of things get set up twice.

Neither is wrong — a real tradeoff between convenience and separation.
4

Technical comfort

not comfortablevery comfortable
anxiouscurious
5

Use-case priorities

6

Logistics for build & orientation