People want one of these. Inbox Quest is built so the first one earns its keep on day one and the other two make you come back.
An AI judge reads each thread in your real inbox and answers the questions that matter: what was asked, whose move it is, how urgent it really is, and whether it is a scam. Then the text is deleted. You get an inbox sorted by what matters and a graded day every evening.
Every verdict becomes narration from a guide who talks to you like a person and tells you when something on your desk has started to smell. Nothing shown to anyone else ever contains a sender, a subject or a client.
One party member per inbox. A month is a journey, a week is a quest, a day is a leg. Good days pay; bad days cost. Never a lost level, never a lost character. And one switch turns the whole game off for the people who just want the tool.
Real screens from the working app, with a fictional inbox. Every name and subject is invented; yours are never shown to anyone.





Same rules, same stats, same honest scoring. The world changes everything you see: the map, the party, the loot, and who's talking in your ear. Click one to retheme the whole site.
The Kingdom ships first. The other three are planned and will be unlocked as skins.
Written as if it's all built. It isn't yet. See the roadmap for what exists today.
Target windows, not promises. Each phase ends with a real test on real inboxes before the next begins.
Updated as the build moves. Last update: September 17, 2026. The founder's two inboxes have run on it every day since September 14.
| Item | Status | Note |
|---|---|---|
| Direct Gmail connect and sync | Done | Every action lands within a minute: received, sent, archived, spam, trashed, read. History up to a year, headers only. |
| The judge | Done | Rules version 4. Kind, ask, whose move, urgency, difficulty, reply and inaction grades, praise, promises, the sender's name, scam with the reason. Reads the other party's history with you before deciding. |
| Scam detection | Done | Sender domain, the real host behind each link, Reply-To and SPF, DKIM and DMARC results all go to the judge. Red skull rows. |
| Review and corrections | Done | Confirm or correct any field; corrections feed the accuracy harness. |
| Inbox, People, Stats | Done | Search, kind filters, sortable columns, working-time ages, ally countdowns, auto prioritize, plan to zero, a year of charts, unsubscribe lists. |
| Goals, medals, the journey | Done | Daily goals, iron to diamond medals, the month as a road, the day as a vial. Morning briefing and weekly digest in the bell. |
| Corporate mode | Done | One switch removes the game. Plain or any world's palette. |
| Billing | Done | Bring your own Anthropic key, or metered at cost plus 10% with a self-set limit. The first read of an inbox never counts. Model picker with honest trade-offs. |
| Owner's admin panel | Done | Who is using it and how much. Counts and dates only. |
| # | Step | How we'll know it works |
|---|---|---|
| 1 | A domain, and the app on it | The site and the app live at a real address |
| 2 | Google verification | Nobody has to reconnect every seven days |
| 3 | Friends beta, Gmail only | Two or three inboxes that are not the founder's, graded honestly for a fortnight |
| 4 | Estimated snoozes | Parked letters stop counting as handled, when you turn it on |
| 5 | The team edition | A manager sees the team's verdicts and subjects, never a body, and the combined inbox works |
| 6 | The guide and the prologue | The game opens as a safe tutorial while it learns your inbox |
The first waves are small and on Gmail. Tell us about your inbox and we'll tell you when a seat opens.
Yes. An AI judge reads each thread to decide what was asked, whose move it is, and how well it went. Then the text is deleted. Only the verdict is kept, and nothing shown in the game ever contains a sender, a subject or a client.
Through our judge to Anthropic, briefly, then gone. You can run the judge on your own Anthropic key so the account it goes through is yours, or later a self-hosted scout so nothing ever touches our servers.
Yes, and it tells them apart from spam. The judge sees who really sent the mail, where every button really points, and whether the message passed its authentication checks. A fake Google notice from a random domain gets a red skull and one sentence saying why. Ordinary junk gets a quiet green tint and a suggestion to unsubscribe.
Yes. A switch in the header turns corporate mode on: the same verdicts and numbers with the game and its language removed. Percentages instead of medals, a plain overview instead of the road. You can keep a world's colours or use a purely functional palette.
Gmail and Google Workspace first. Outlook and IMAP after the beta. Any number of inboxes per person; each one is a party member.
The beta is free apart from the AI. For that you either bring your own Anthropic key and pay them directly, or use ours and pay exactly what the judge cost plus ten percent, up to a monthly limit you set. A busy inbox runs about a dollar a day on the recommended model; the first read of an inbox is about a cent a thread and never counts against your limit. After the beta, a subscription on top. Heavy users never subsidise anyone and nobody subsidises them.
Not today: there are no team accounts yet. When the team edition arrives, a manager sees the judge's verdict and the subject line for each thread, never the body, never your corrections, never the judge's notes on your correspondents. You are told that in one sentence before you join a team.
Both, and a story. First it is a triage tool: the judge tells you what needs you and grades your day. Then it is a LitRPG narrator that turns those verdicts into a story. Then it is a game with stats, loot and a party. You can use the first without caring about the other two.
You can lose a fight. You never lose levels or gold. You retreat, heal, and come back. A bad week is redeemable. That's the whole point.
The fastest way to reach us right now is the beta request form. There's a box for what you dread about your inbox; it doubles as a mailbox for anything else. A proper contact address arrives with the domain.
Inbox Quest works by reading your email. That deserves a plain answer about what happens to it. Last updated September 17, 2026. This policy applies to the closed beta; it will be updated before public launch.
When you connect a Gmail account, you grant Inbox Quest read-only access to your mailbox (the Gmail read-only scope) and your email address. We use it to list messages, read message headers, and read the text of threads that need a verdict. By default we never send email, change labels, delete mail, or write anything to your mailbox. If you turn on label writing in Settings (off by default, and it asks you to reconnect with Google's modify permission), Inbox Quest adds and removes three labels of its own on your threads (Inbox Quest/Waiting on you, Inbox Quest/Urgent, Inbox Quest/Ally) and still never sends, deletes or moves mail or touches labels it did not create.
Message metadata: sender and recipient addresses, subject, date, thread id, and the actions you took (received, sent, archived, marked spam, read). The judge's verdict for each thread: what was asked, whose move it is, urgency, difficulty, whether you promised something, the sender's name as they signed it, and for a scam the reason. Your settings: working hours, timezone, world, class, goals and view. Daily counts built from the above. If you bring your own Anthropic key it is stored encrypted, shown afterwards only by its last four characters, and used for nothing but your own verdicts.
Email body text is never written to our database. It is fetched, sent to the AI judge, and discarded. A short encrypted cache of a thread's text may exist for up to your chosen retention window (7 to 90 days) so the judge does not re-read a thread every time it changes, then it is deleted. Only the verdict stays.
Thread text is sent to our AI provider (Anthropic) to produce the verdict, under terms that forbid training on it. Nobody at Inbox Quest reads your email. Friends who view your party see stats, items and story, never senders, subjects, health or performance. If you turn on developer mode, you can see the prompt and raw response for your own threads; nobody else can. Team accounts do not exist yet; when they do, a manager will see verdicts and subject lines only, never bodies, and members will be told so before joining.
OAuth tokens are encrypted at rest with a key that lives only on our servers. You can disconnect an inbox at any time from Settings, or revoke access at myaccount.google.com/permissions. Disconnecting deletes the token immediately; metadata and verdicts follow within 30 days, or at once on request.
Inbox Quest's use of information received from Google APIs adheres to the Google API Services User Data Policy, including the Limited Use requirements. Google user data is used only to provide the features described here, is never sold, never used for advertising, and never used to train general AI models.
To delete your account and everything attached to it, use Settings or write to us through the beta form. We respond within a few days during the beta. A proper contact address arrives with the domain.
Sign-in arrives with the closed beta. Request a seat and you'll get a link when yours is ready.