One memory. Every model.
Rent the intelligence. Own your memory.
Whether you use one AI or five, they forget you - or worse, what they know about you belongs to them. Virgil is one memory underneath every AI you use - ChatGPT, Claude, Grok, whatever comes next - so each one picks up exactly where you left off, and the record stays yours.
Connect it once and keep using the tools you already use. The intelligence you rent gets replaced every few weeks. The memory you own compounds. Virgil holds the part that’s yours - and if you ever want out, the whole record exports in an open format and goes with you.
Tell ChatGPT once. Claude already knows - see the live demo.
Your brain, unlocked.
90 seconds. Three platforms. Pick up right where you left off.
One uncut take, clock on screen: Claude saves a fact, ChatGPT recalls it and adds to it, Grok answers with the whole story - all from one record that belongs to you.
Every AI you run knows you. You own none of it.
Give the labs their due - platform memory got real this year. ChatGPT knows you. Claude knows you. Separately. Each memory works only where it was captured, and it works for them - it’s the reason leaving feels expensive. You can export a pile of transcripts, sure. What moves is a snapshot - it lands once and stops keeping up. So every switch, you’re the bridge: re-briefing the new model on things you already settled somewhere else. All that intelligence on retainer, and you’re still the courier. That’s the wall. Virgil takes it down.
Don’t rent your own brain back.
Two clocks run here. On the short one, intelligence is a rental - a new best model ships every few weeks, and you should jump the day it does. No loyalty. On the long one, your context accrues - years of decisions, corrections, original thought. Scattered across platforms, it’s their asset, rented back to you at the price of staying. Virgil puts it on your side of the table: one record of plain facts - dated, sourced, corrected in place - compounding for you. A record is only yours if you can read it, fix it, and take it with you. With Virgil you can do all three.
Not switching. Tuning.
Run one AI or run five - Virgil is the record underneath, so however you split the work, it stays one conversation.
- The right model for each job - and change your mind tomorrow without re-introducing yourself.
- Day one on every frontier release - your full context is already there the morning it ships. Keep the new model or drop it; your record doesn’t care.
- Cheap work on cheap models - stop paying premium prices to summarize a document.
- One record under all of it - your stack stops being five strangers and starts being one machine.
No lock-in. Not even ours.
Virgil doesn’t wire your models together - each one connects to your memory, and none of them to each other. Drop a model, add a model - nothing breaks when you walk away from a platform. That includes walking away from us. Your whole record exports anytime in memvelope, an open format we steward in the open - the spec and source live at github.com/memvelope/memvelope, and your record is plain JSON that reads without Virgil - any text editor opens it. If it ever stops earning its keep, you leave with everything. Ownership you can’t test at the exit isn’t ownership.
No new app. No new habit. No copy-paste.
Connect Virgil once, then live exactly how you already do. You talk; it keeps what matters and hands it back when you need it. Nothing to file, nothing to tag, nothing to babysit. No second brain to feed on Sunday nights. You don’t manage the record. You just own it.
Setup is a two-minute connection inside the apps you already use - ChatGPT, Claude, Grok - using the same connector standard (MCP) the industry runs on. See the setup guide.
Already connected? Bring what your other AI already knows about you into Virgil, once: fill your record.
Not a promise. A procedure.
Three moments from the live product, exactly as it behaves:
Print deadline is Friday, July 17.
Claude - later that day, nothing re-explainedYour print deadline is Friday, July 17.
you said - 2026-07-06The answer arrives with its receipt: who said it, and when.
inserted 1duplicate 0one fact_id
Same fact, saved again from the other modelinserted 0duplicate 1same fact_id
Virgil dedups server-side. One record, no copies - these counters are the demo protocol’s actual output.
- Ask for your export in chat.
- A one-time code lands in your email; your passphrase seals the pickup.
- The whole record leaves as memvelope - plain JSON, readable without Virgil in any text editor.
The exit is a shipped mechanic, not a policy promise.
Faithful illustrations of the live product’s output - labels, dates, counters, and ceremony as shipped. The real thing, run live, is at /demo - one command reproduces the whole protocol, receipts included.
Open the hood. Read a fact.
Every fact Virgil keeps carries three tags: the date it landed, the AI it came through, and whether you said it or Virgil inferred it - [you said] and [inferred], never blurred. When something changes, the new fact supersedes the old one - the old one stays underneath, in history, so you can trace what changed and when. Here’s one fact, as it actually lives in the record:
Maya's our new head of sales.
Chat apps are one wall. Agent workspaces are another.
Virgil works inside chat apps today: ChatGPT, Claude, Grok. Buzz is not a chat app. It is an open workspace where people and agents share channels.
We built a Buzz connector and ran it against an export of our test workspace on 2026-07-24. That export became date-stamped pages in a Virgil record. A keyword search then found a March pricing discussion and quoted it back with the author and the #pricing source line.
Your record moves as data.
When Virgil stores a fact or hands one back, no AI reads it - on our side it moves like a record in a filing cabinet, not a conversation. Saving and recall stay zero-LLM: nothing on those paths can misremember, embellish, or train on what you said. The guided paste keeps the same posture - it’s a one-time paste, what you paste lands as dated [you said] facts you vouched for, and no model of ours reads it. There is one exception and we’ll name it: drop your full ChatGPT or Claude export instead and a small distiller model turns that conversation text into dated facts labeled [inferred]. That is the one place an LLM on our side runs, at import only.
Your AI chat is the most expensive place to keep your memory.
The real cost is re-explaining yourself - every new chat, every new model, every month, forever. The chat window makes you hold your own context: keep one long thread alive until it slows down and hits the wall, or open a fresh one and brief it from scratch. Native memory stays with the assistant that captured it - you can copy a snapshot out by hand, but the copy is frozen the moment you make it. Virgil moves your record outside the thread: start fresh anywhere, pull only the slice you need, and every model you run picks up where the last one left off. If you pay per token, the redundant history you re-send every turn is real money on its own - in that case Virgil pays for itself on the token bill alone. For everyone else, the payback is time: the briefing you never give again.
A guide who forgets the way is just company. Virgil knows.
Stop re-explaining. Start getting sh*t done.
Introductory pricing - the price you join at is the price you keep. Virgil Core is the record - plus your morning brief, every day in your inbox: $99 a year - under $2 a week - or $18 a month. Virgil Pro adds client-confidentiality lenses and meeting briefs: $249 a year or $29 a month. Either way the record is yours - read it, correct it, export it, cancel and keep it. One shared context makes every model you run sharper than it is alone. The intelligence will keep changing. Your memory shouldn’t start over every time it does.
Virgil Core
The record itself: save, recall, correct, import, export - one memory across every model you run. Plus a morning brief from your own record, daily.
Virgil Pro
Everything in Core, plus the professional layer: client-confidentiality lenses that wall off what must stay separate, and meeting briefs built from what you already told it.
Sign-ups are paused right now. We bring records up carefully - in small groups, never in a crush - and the current group is full. Leave your email and you’re first in line when the doors reopen.
30-day money-back guarantee - and your record exports either way.
Checkout by Stripe · Cancel anytime · Two-minute setup after checkout
Which AIs work today? ChatGPT, Claude, and Grok - the three we stand behind today.
What’s Core vs Pro? Core is the whole record - save, recall, correct, import, export. Pro adds the client moves: lenses and meeting briefs. The morning brief comes with Core.
What if you shut down? Your record is yours - export it anytime, it opens in any text editor.
Is it private? Saving and recall are zero-LLM - no AI reads your record. Only a full-export import runs a distiller, once.