Ask, do not browse
Describe the person, time, topic, or loose memory you have. MailCat resolves the people and thread, then returns the coherent exchange—not a page of keyword matches.
AppForCat concept 08 / MailCat
No folders. No labels. Just ask.
MailCat is email rebuilt around an adaptive agent. Tell it what you need; correct it when it misses; and it learns your language, priorities, and boundaries until only what matters breaks through.
Askin natural language
Focuswhat matters now
it learns / you decide
The inbox should do the organizing. You should do the deciding.
Email still hands you a filing cabinet. Every new message asks you to scan, sort, label, archive, search, and remember. Categories change the drawer names; they do not remove the cognitive work.
MailCat replaces navigation with conversation. Ask for the exchange with Jane last week. Ask what needs an answer today. Ask it to draft a reply with the facts already sitting elsewhere in your mail.
Under the idea is a real mail service: AppForCat addresses receive into a private inbox, and products send authenticated mail through Resend. The agent is not floating above a demo corpus; it is being designed for a mailbox that can actually send and receive.
Three moves + one memory
Find. Write. Tend. Without managing a filing cabinet.
The same conversational surface handles the whole loop. What changes is your intent—not which tab, label, or mode you remembered to open.
Draft with the full context
MailCat brings the relevant thread, facts, and your developing voice into the draft. The agent proposes; you remain the final author and press send.
Teach it what is noise
Point at what you do not want. MailCat shows the rule it inferred before applying it, keeps suppression reversible, and can still retrieve muted mail when you ask.
Make every correction count
Your choices become durable personal signals. “Jane” resolves correctly next time; relevance reflects your work; the agent becomes sharper without forcing you to maintain a taxonomy.
The learning loop
From inbox flood to a focused beam of truth.
Not one universal ranking. A relevance model grounded in your language, your corrections, and what you are trying to do right now.
Mail arrives as real life does
Threads, requests, newsletters, receipts, collaborators, and noise enter through the live inbox.
The agent forms a useful view
It connects people, time, threads, asks, and meaning to retrieve or suggest what seems relevant.
You choose, correct, or redirect
Pick the right Jane, reject a noise rule, refine the question, edit the draft, or approve the send.
The next pass gets sharper
The resolution and preference become a weighted memory the agent consults the next time.
The contract is simple: the AI proposes; the human disposes. Nothing sends, deletes, or generalizes into a lasting rule without your say.
What changes
The inbox stops being a place you manage—and becomes a question you ask.
Taxonomy disappears
No folder tree to design, label scheme to remember, or category model to fight. Natural language becomes the interface to the mailbox.
Relevance becomes personal
“Important” is not a global score. It is learned from what you open, correct, mute, retrieve, draft, and choose to act on.
Answers remain grounded
The focused result carries the messages and thread evidence behind it. The beam is narrow because it is precise—not because the rest was destroyed.
What is real today
The AppForCat receiving service, private inbox, and Resend outbound path are live. The full adaptive agent experience is the working product concept under development. It does not depend on copying years of old mail: it can begin with the live inbox, learn from the first correction, and improve forward.
Live mail below.
An adaptive agent above.
See the working inbound and Resend infrastructure, then examine the proposed retrieval, memory, learning, and human-control seams.
Open the technical details