Virgil is your memory across the AI you already use. He keeps a dated, sourced record of your world - people, decisions, preferences, what's changed - and hands the right piece back the moment you need it, inside Claude, ChatGPT, and Grok.

No. You connect Virgil once inside the AI you already use, then just talk - nothing to file, tag, or manage.

Claude, ChatGPT, and Grok today, through a custom connector you add once - the same record travels across all of them. Tell one, and it's already waiting in the others.

Yes - Grok connects the same way Claude and ChatGPT do, and the import can also fold in what Grok already knows about you. Virgil speaks an open standard that more AIs adopt every month; Claude, ChatGPT, and Grok are the three we stand behind today. We'd rather promise you three that work than five that mostly do.

Not officially yet. Virgil speaks MCP - the same open standard Claude Code uses for its connectors - so the door is technically there, but we haven't walked that path end to end and won't claim it until we have. If you try it and something snags, reply to any Virgil email; we'd genuinely like to hear.

Yes. MCP is the open standard Claude and ChatGPT both use for connectors - it's why one Virgil works in both apps with nothing to install, and why he isn't chained to either. You paste one address, sign in once, and he's there.

Two ways in, both on the import page. The guided paste, about ten minutes: Virgil hands you a short prompt; you run it in Claude, ChatGPT, or Grok - each answers with the durable facts it knows about you - and you paste those answers back; everything you paste lands dated and labeled [you said]. Or drop your full ChatGPT or Claude export: the file is read on your device and never uploaded - only conversation text reaches Virgil, where a small distiller model turns it into dated facts labeled [inferred], held at lower confidence, with the dates your history actually carries. The distiller runs at import only - saving and recall stay zero-LLM. Either way, come back and add more anytime; he won't duplicate what he already holds.

The things you'd want back: people, commitments, dates, prices, decisions, preferences - the dinner that can't move, not the small talk around it. Every fact lands with its source and the day it landed. He treats uncertain things carefully, and he flat refuses to keep card numbers, government IDs, and passwords - some things don't belong in a record.

You never have to learn a tool name - you just talk, and Virgil picks the right one. Saving and recall, timelines, corrections that keep their history, a memory map, an export that leaves with you, even a proof you can run yourself. The full shelf, with what to say for each, lives at virgilknows.com/docs/your-tools.

Careless memory does - stuffing everything you've ever said back into the AI crowds out the question you actually asked. Virgil's whole job is curation: keep what matters, drop the noise, hand back only the piece the moment calls for. We measure how well he decides, and tune it - a memory that hoards everything is as useless as one that keeps nothing.

ChatGPT's memory is genuinely useful, and it lives inside ChatGPT. It doesn't follow you to Claude, and its shape is OpenAI's to set - the June 2026 update made a system-written summary the default and demoted the list you used to curate to a legacy setting. Virgil is a record you own outside any one AI: dated, sourced, the same in every chat, exportable the day you want it.

Same trade, other direction. Claude's memory is good - and it lives in Claude. It doesn't travel to ChatGPT, and it isn't a record you can hold in your hands. Virgil keeps one dated, sourced record outside both, so switching models never costs you your history.

For you - nothing. Your record doesn't live inside either one, so a vendor redesign can't migrate it, prune it, or sunset it. That's most of the point: your memories shouldn't die because someone else redesigned memory. If the way connectors plug in ever changes, that's ours to chase - the record stays put.

It half-solves it, once. Claude ships an import (labeled experimental) that takes a summary you paste over from your other AI, and ChatGPT's memory summary is text you can select and copy out by hand - both are one-time moves of a snapshot, and what arrives is prose a model wrote, not your dated record. Nothing stays in sync afterward. The labs shipping these doors is the concession that memory should move. Virgil is the version where it never has to: one record connected to Claude, ChatGPT, and Grok, current everywhere, every fact labeled and dated.

Yes. The record is yours alone: every fact carries where it came from and when, nothing is sold, and you can export it or have it deleted anytime. You pay for Virgil - that's the whole business. There's no ad model and no data business behind it.

At the database level, not just in the app. Your record is sealed to you where it's stored - the database itself refuses to show your rows to anyone else, even if a bug upstream were to ask. Each record also gets its own keys, so one can never unlock another. It's the structural kind of privacy - the kind that doesn't depend on anyone remembering to check.

No. Virgil trains no models on your record, and nothing you store is sold or shared. Your chats run on your own Claude or ChatGPT account, under the terms you already have with them - Virgil just hands them the right piece of your record when they ask.

Yes - tell him to forget it. He retires the fact on the spot: dated, no longer surfaced, done. The record keeps a dated trail of what changed rather than pretending it never happened - that's what makes it auditable - and deleted notes stay recoverable for 72 hours in case you change your mind. Want everything gone? Reply to any Virgil email and we'll close the record completely.

Yes - ask him in chat, and the whole record goes with you: every fact with its source, date, and revision history, in memvelope - an open, documented format anyone can read without Virgil. We maintain the format in the open at github.com/memvelope/memvelope, and your export opens with any text editor - plain JSON, no Virgil software - ownership you can test, not take on faith. Pickup takes a one-time code to your email plus a passphrase only you know - your whole record deserves more than one chat session's say-so.

The format your export ships in: one plain-JSON envelope holding your record - facts, dates, sources, said-vs-inferred labels, and the history of corrections. It is MIT-licensed with a public spec at memvelope.com, so it reads without Virgil - any text editor, any JSON library. Honest scope: a young format we steward in the open, not an industry standard. The guarantee is structural, not social.

Then you leave with everything. The full record - every fact, source, date, and revision - exports in memvelope, an open, documented format that reads without us, anytime, including the day we turn the lights off. A memory company that could hold your memories hostage would have built the wrong product. Export-anytime is the structural answer, not a promise.

$18/mo, or $99/yr paid annually - about 54% off if you pay for the year. 30-day money-back guarantee, no questions asked. Cancel anytime.

Because a sign-up here isn't a login, it's a build: we provision your own isolated record and pre-populate it with what your AIs already know about you - real cost on our side from day one, not a demo tier. Instead of a watered-down free version there's the 30-day money-back guarantee: if he isn't for you, full refund, no questions asked. The risk sits with us, not you.

Not yet. Virgil keeps one person's record, sealed to that person alone - so a household means one subscription each today. If a family plan matters to you, reply to any Virgil email and say so - it genuinely shapes what gets built next.

Reply to any Virgil email and we'll take care of it - no forms, no retention script. Inside your first 30 days, that includes the full refund, no questions asked. And the record stays yours: export it before you go, in a format you keep.