Categories that get your work.
Jev reads the context to suggest where a message belongs. Customer conversations, build alerts, and finance mail get their own space.
Less noise.
More “this matters.”
Business email, organized with Jev by TypeSafe.
Find the customer who needs you. Give every message a place.
See the signals behind what gets your attention.
Sign in or create a MailKite account, then return to your inbox.
TESTING BETAJev sorts messages when you open them. We’re learning as we build.
MailKite / Smart InboxSample preview
LESS NOISE. MORE MOMENTUM.
Your inbox, with intention.
One message. Separate signals for category, attention, and spam.
Fictional messages · Illustrative scores · Select a message to explore
Inside the sortJev
Illustrative judgment
Our team is blocked on checkout
Hi! Our team is seeing an error on checkout. We’re launching this afternoon and can’t complete a test purchase. Could you help us get unblocked?
Main categoryCustomers
Needs timely attention96%
Spam probability2%
A customer issue with a near-term deadline. In this example, the attention signal puts it in Important as well as Customers.
Inside the sortJev
Illustrative judgment
Production deploy completed
Deployment complete. All checks passed and the release is live in production. No action is required. Have a great day shipping!
Main categoryBuild
Needs timely attention12%
Spam probability1%
Engineering mail belongs in Build. A successful deployment can be relevant without needing timely attention.
Inside the sortJev
Illustrative judgment
Your September receipt
Thanks! Your September subscription payment has been received. This receipt is for your records. There is no outstanding balance or action required.
Main categoryFinance
Needs timely attention8%
Spam probability1%
A routine receipt belongs in Finance. Its low attention estimate keeps it out of the Important view.
Inside the sortJev
Illustrative judgment
Claim your guaranteed cash prize
You have been selected for a guaranteed cash prize. Send a processing payment immediately to claim money from a competition you never entered.
Main categoryUpdates
Needs timely attention3%
Spam probability98%
A high spam estimate puts this example in Spam review, rather than ordinary category views. It remains visible in All inbox.
For founders, developers & marketers Focused AI decisions Visible probabilities Your rules, in plain English
A LITTLE COMMON SENSE FOR YOUR INBOXEvery email has a place.
Every email has a place.
Not every email needs you now.
A customer question, a deploy notification, and a receipt shouldn’t compete for the same attention. Jev helps tell them apart.
Importance is a probability.
See an estimate of whether a message needs timely attention. Important brings attention-needed mail together across categories.
Needs timely attention96%
Illustrative estimate, not a guarantee
Smart spam review.
Jev considers the message alongside delivery signals. High spam estimates move evaluated mail into a separate review view, still visible in All inbox.
Reviewable. Not silently deleted.
Primary
Conversations, coordination, and the people behind the work.
Customers
Support requests, new leads, demos, and honest feedback.
Build
Pull requests, deployments, and the alerts that keep you shipping.
Growth
Partnerships, campaigns, and newsletters worth a look.
Finance
Invoices, payments, receipts, and subscription updates.
Updates
Routine notifications, all with a place of their own.
WHY WE’RE BUILDING WITH JEVSmall decisions.
Small decisions.
Useful intelligence.
Jev is TypeSafe’s first System One model: built to give software focused, typed decisions and probabilities. That makes it a natural fit for the repeated judgments inside an inbox.
The judgment is AI.
The rules stay explicit.
Instead of asking for a paragraph and parsing an answer, we ask specific questions: What category fits? Is this likely spam? Does it need timely attention?
Jev returns structured signals. MailKite uses them to decide which views show the message. You can inspect the estimates instead of trusting an unexplained label.
Explore TypeSafe’s documentation THE SHAPE OF A DECISIONIllustrative
01Message + contextMailKite
ChoiceWhich category fits best?NoulDoes this need timely attention?NoulIs this spam?
03Inbox views + review rulesYour app
One category decision. Independent attention and spam probabilities. Questions evaluated together.
For the curious: probability isn’t the same as confidence
A Choice returns one category, probabilities across the available categories, and confidence describing how concentrated that distribution is. A Noul returns the probability of “yes”—such as whether a message needs timely attention—with no separate confidence field. Typed output makes integration predictable; it does not make every judgment correct. Our beta thresholds still need evaluation on representative mail. Read TypeSafe’s explanation ↗
YOUR PROJECTS. YOUR RULES.An inbox that speaks
An inbox that speaks
your language.
Make a tab for your fundraise. Or a product launch. Describe what belongs there, and Jev checks that rule alongside the standard categories.
One message can match several custom tabs. Up to eight rules, written in plain English.
Try your own rule in the appA tab for what you’re buildingExample
TAB NAMEFundraising
WHAT BELONGS HERE?Investor conversations about our fundraising round, introductions, or diligence requests. Exclude newsletters and vendor pitches.
A rule example, not a live evaluation.
ROADMAP · NOT IN THIS BETAJev decides.
Jev decides.
An LLM could take it further.
Let focused models handle triage. Bring in a generative model when a message needs a draft, a summary, or deeper reasoning.
We’re exploring automatic escalation: route a time-sensitive customer issue to a person, or an uncertain case to a stronger model, under explicit rules.
The pattern behind the ideaA POSSIBLE NEXT STEP
Incoming messageContent + business context
Jev evaluates the signalsIntent, attention, uncertainty
OrganizeLLM assistHuman review
An LLM could draft. Jev could check specific claims against the source. Code would decide when to revise or ask a person.
BUILDING IN THE OPEN
Source release in preparationA small app. An idea you can build on.
We’re preparing Smart Inbox for an open-source release, so you can explore how Jev’s decisions fit into a real email app. Public repository and license coming with the source release.
A FEW GOOD QUESTIONSBefore you
Before you
open your inbox.
It’s a beta. Here’s where things stand.
TypeSafe models & data handlingWhat does Jev actually do?
Jev is TypeSafe’s System One decision model. In Smart Inbox, it selects a category and returns independent probabilities for spam, timely attention, and your custom rules. You can inspect these signals in the app. They are estimates, not guarantees.
Does it sort everything automatically?
In this testing beta, Jev evaluates a message when you open it, or when you explicitly re-evaluate loaded message bodies. The resulting signals automatically determine its inbox views. Whole-mailbox background sorting is not available; search and counts cover loaded messages.
Will signup bring me back to Smart Inbox?
Yes—the Continue with MailKite link starts Smart Inbox’s sign-in flow. Choose Sign up on MailKite if you need an account, then complete authorization to return to the inbox. Already have a session? Open the app directly. A new account may not have mail yet; the app also offers a sample inbox.
Can I connect Gmail or Outlook?
This beta uses mail available to your MailKite account. Signing in with Google verifies your identity; it does not connect your Gmail mailbox. External Gmail and Outlook mailbox connectors are not available in this beta.
Does it write replies or escalate messages?
Automatic escalation and LLM-assisted drafting are roadmap ideas. Today, the beta provides sorting and attention signals. Its assistant tools are labeled mock previews, not live LLM output. Jev itself returns decisions rather than generated prose.
Where does my email go for analysis?
When you evaluate a message, its content is sent server-to-server to TypeSafe. Smart Inbox keeps MailKite OAuth tokens on the server; the browser receives a session cookie. TypeSafe states that customer requests and responses are not used to train Jev. See the linked TypeSafe model documentation for its data-handling details.
Can I make my own categories?
Yes. Add up to eight custom tabs with plain-language rules. A message can match several rules while retaining its main category. Tab definitions stay in your browser for now; cross-device sync, learning from corrections, and MailKite label writeback are not available yet.
Where can I find the source?
We’re preparing Smart Inbox for an open-source release. The public application repository and license will be linked here when published. Jev is a separate model served by TypeSafe.
Give your attention
a better starting point.
Try Jev on your MailKite mail. Open a message.
Inspect the signals. Make the inbox yours.
Just exploring? Open the app and choose “Explore sample inbox.”