Practical

Magic Inbox: how forwarding an email becomes memory

Forward any email to AI Twin and it becomes typed, searchable memory. Here is how Magic Inbox works, and where its limits are.

By AI Twin · 15 July 2026 · 4-minute read

Some of the most important context in your life arrives by email and then disappears into an inbox you will never search again. The booking confirmation, the quote, the contract, the appointment letter, the thread where something was agreed. It is all there, and none of it is findable when you need it.

Magic Inbox is our answer to that. Here is how it works, and where it stops, because the limits matter as much as the capability.

The mechanic

When you set up AI Twin, you get a unique forwarding address at inbox.ai-twin.app. Forward any email to it, or set up a rule that forwards certain emails automatically, and AI Twin does three things.

First, it reads the email and any attachments. Not just the body text, the structure: who it is from, when it arrived, what it contains, what kind of document is attached.

Second, it classifies what it found and turns it into typed memory. A booking confirmation becomes an event with a date. A quote becomes a document linked to the company that sent it. A letter becomes a record you can find by meaning or by name later.

Third, it files it with provenance. The memory knows it came from an email, when, and from whom. You can see that trail. Nothing is stored as a mystery.

After that, the email is findable the way all your memory is findable in AI Twin: by meaning, by exact name, or in conversation. You do not have to remember which folder it went in, because it did not go in a folder. It became memory.

Where it works, and where it does not

We grade our capabilities honestly, so here is the honest picture.

The document reading is strong. Word, PDF, calendar invites, spreadsheets, the common formats that email carries all parse cleanly. Attachments in the world’s major languages read well, because the reading is done by strong vision and language models. Extremely unusual scripts or very low-quality scans are the soft spot, and we say so.

The classification is good and getting better. AI Twin is careful rather than clever: it would rather file something plainly than guess wrongly. If it is unsure, it holds the item for your review rather than acting silently. Consent-first means the system waits for you, and Magic Inbox is no exception.

What it is not: it is not reading your whole inbox. It only sees what you forward or what you set a rule to forward. It does not connect to your email account and scan everything. That is a deliberate boundary. Your inbox is yours; you decide what becomes memory.

Why we built it this way

We could have built an integration that reads your entire mailbox. It would capture more. It would also be exactly the kind of always-on, watch-everything design we do not believe in.

Forwarding is a small act of consent. You choose the email. You send it. It becomes memory. That is the model: you in charge, the system preparing, nothing happening that you did not ask for.

Magic Inbox is ready at launch. If you use it and something does not classify the way you expected, tell us on our feedback page. That kind of report is exactly how we make the classification better.

Share this post

Newsletter

Join the newsletter

AI Twin is built and sign-in opens shortly. We will tell you when.

Newsletter

Notes from the build.

AI Twin is built and sign-in opens shortly. We will tell you when.

More from the Journal