How to Organize Important Documents With an AI Agent
A document vault your agent can actually use: what to scan, how to tier by sensitivity, what it may read, and the renewal calendar that stops surprises.
A folder full of PDFs is storage; a document vault is a system an agent can actually answer questions from. The difference is three things you decide up front: what is in it, how each file is tiered by sensitivity, and what your agent is allowed to do with each tier.
Get those right and "where's the policy number?" stops being a search and becomes a question. This is the setup, in one evening.
Why a folder is not a system
Most people's important documents live across four places: a paper drawer, a downloads folder, an email inbox where the PDF arrived, and a cloud drive nobody opens. Nothing is missing, but nothing is findable either — because there is no naming convention, no tier, and no expiry date attached to anything.
Your agent cannot fix that by reading harder. It needs structure it can trust. Three properties make a vault agent-readable:
- Consistent names. One pattern, no exceptions, so a search is deterministic.
- Explicit sensitivity. Not every file deserves the same access.
- Dates that mean something. Issue date, expiry date, and the date you last looked at it.
The three tiers
Tier your documents before you hand anything over. This is the single most important decision in the whole setup.
| Tier | What goes in it | Agent access | Example |
|---|---|---|---|
| 1 — Public | Anything already public or low-risk | Free read, free quote | Warranty PDFs, manuals, old receipts, utility tariffs |
| 2 — Private | Identifies you or your household | Read on request, quote only when relevant | Policies, contracts, leases, service records, vehicle papers |
| 3 — Sealed | A leak causes real financial or legal harm | Agent may know it exists; never opens it unattended | Passports and IDs, bank and card statements, medical records, tax filings |
Tier 3 files still belong in the vault — they just live in an encrypted area your agent indexes by name and expiry date only. The FTC's consumer guidance on which documents to keep and which to shred is a good reference for what lands here, and its identity theft advice is blunt about the reason: financial records, Social Security and Medicare cards and anything with personal information on it needs a safe place and shredding when you are done with it.
What to scan first
Do not try to digitise the drawer in one weekend. Work in this order:
- Anything with an expiry date. Passports, IDs, visas, licences, registrations, insurance policies. These are the documents that cause real pain when they surprise you.
- Anything you have needed in the last 12 months and had to dig for. That is your actual high-value list, not a theoretical one.
- Anything you would need in an emergency, away from home. Insurance policy numbers, emergency contacts, medical details.
- Anything a bank, lawyer or government office has asked you to produce. If it has been requested once, it will be requested again.
- Warranties and purchases still in their window. Cheap to file, easy to forget, and worth real money when something breaks.
Everything else can wait. Old statements, closed accounts, statements you have already reconciled — ideally those go in the shredder, not the vault.
Naming, so the agent can actually find it
Pick one convention and never break it. This works:
YYYY-MM-DD-category-issuer-short-description.pdf
2026-03-14-insurance-axa-home-policy.pdf
2029-08-02-identity-passport-younes.pdf
Three rules that make the difference:
- Dates first and always ISO. The document's issue date, not the day you scanned it, so sorting by name sorts chronologically.
- Category, then issuer, then subject. You will search by one of those three, every time.
- Never "final", "v2" or "scan_001". Your future self and your agent both need to know what the file is without opening it.
Then add one index file — a plain table with columns for name, tier, issuer, issue date, expiry date and location. It is the cheapest thing you will build and the one your agent will read most. If you want the reasoning behind that, this is how agent memory should be structured.
The renewal calendar
This is where the vault starts paying for itself. Every document with an expiry date becomes a calendar entry, filed with a reminder far enough out to act:
| Document | Remind me | Why |
|---|---|---|
| Passport | 9 months before expiry | Many countries refuse entry inside six months |
| Visa or residence permit | 4 months | Processing queues |
| Vehicle registration | 60 days | Inspections and payment windows |
| Home and health insurance | 45 days | You want time to compare before renewal, not after |
| Warranty / guarantee | 30 days before it lapses | Claim before it expires, not after |
| Tax filing | 90 days | Documents take longer to gather than you think |
The IRS is a useful anchor for how long to keep the tax paper behind it: generally three years from the date you filed, and longer for property, stock transactions and anything where the period of limitations has not run out. Put the retention rule in the index file so your agent can tell you what is safe to destroy instead of hoarding everything forever.
What your agent may and may not do
Write this down before the first read, not after the first mistake:
- May: read Tier 1 freely, search Tier 2 on request, quote a Tier 2 detail when it is directly relevant, and tell you when something expires.
- May not: open Tier 3 unattended, send any document to any third party, upload a document to an external service, or paste a full Tier 3 document into a chat.
- Always: ask before sharing, and log every file it opened.
That last one is not paranoia — it is what makes the vault auditable. If you cannot reconstruct which document your agent read and when, you cannot use it for anything that matters. The audit trail guide covers the mechanics.
And keep a copy. A vault that exists in one place is a single point of failure, and the same discipline that protects your agent's memory applies to your documents.
The document tier picker: choose a document type and see which tier it belongs in, what your agent may do, and how long to keep it
Takeaway
Spend an evening on three things: tier every document by sensitivity, rename so a search is deterministic, and turn every expiry date into a reminder you will actually see. Then your agent can answer the questions you currently answer by emptying a drawer — and the renewal that used to ambush you arrives as a calendar entry with time to spare. Start with the vault folder in Wolffish's start guide, and pair this with the recurring side: the life admin guide covers the tasks that repeat, this covers the documents they depend on.
