Four inboxes,
one mind,
no peace.

I have four mailboxes open right now.

There is the Gmail I have been using since school. There is the second Gmail my college issued me at Rashtriya Raksha University. There is the Outlook account Deloitte gave me for the Cyber Strategy internship. And there is the Zoho workspace we set up for AstraQ Cyber Defence, which is actually two addresses I have to monitor in parallel. One is mine, raj.pundkar@astraqcyberdefence.com. The other is contact@, the one customers and partners write to.

This post is the long version of why we built Metis Mail.

The shape of the problem

Four mailboxes does not sound like a crisis until you live in it for a year.

The first month, you keep them in four browser tabs and tell yourself it is fine. The second month, you start missing replies, because the tab you were not looking at had the thread you were waiting on. The third month, you write a small script that polls all four and pings you on Slack, because at this point Slack has become the thing you actually live in. By the sixth month, you have given up on systems entirely, and you are just refreshing whichever inbox somebody messages you about that morning.

The part nobody warned me about was the AstraQ split. raj.pundkar@ is mine. contact@ is the company. As the team grew, contact@ stopped being something I should be reading alone. It is where pilot enquiries land. It is where a customer's first message about a renewal lands. It is where a security report from a researcher lands. The right person to act on each one of those is not always me, and the worst possible outcome is that mail sitting unread because the only person with the inbox is asleep or on a flight.

What every email client I tried wanted me to do was forward, or auto-forward with a rule, or set up a Google Group, or hand out the password to a junior engineer and hope nobody screws it up. None of those is a workspace. None of those is a system. They are duct tape with extra steps.

This is the standard shape of inbox life for anyone who works in more than one place. A student with two college accounts and a personal one. A consultant with a client mailbox and their own. A founder with a personal mailbox and a company mailbox that the whole team needs to live inside. There are millions of us. None of the popular email clients are built for us.

The default answer the industry has settled into is that you should have less email. That is not an answer. That is a way of saying the customer is the problem. I have customers who write long replies because they are taking the question seriously. I have a team that ships at night and writes about it in the morning. I have a Deloitte manager who lives entirely inside Outlook for compliance reasons. None of these people are going to stop emailing me. The job is to handle the volume. The job is not to wish it away.

The tools that exist did not fit

I tried the obvious things first.

Superhuman is the closest thing to a real answer. It is built for keyboard people. The triage flow is fast. The shortcuts are good. The problem is that Superhuman is single account. If you want it across four mailboxes, you pay per mailbox, you still flip between separate windows, and there is no concept of a shared mailbox at all. Their unified inbox is a list view, not a workspace.

Hey has a strong opinion about how email should work. I admire the opinion. The opinion does not match how I work. It is also single account, and the price scales the same way. Spike, Newton, Mailbird, all of these are single account, or they duct tape a unified view on top of three IMAP connections that fall out of sync constantly.

Then there are the platform clients. Gmail and Outlook are excellent at being their own provider, and average at everything else. Apple Mail is a museum piece. Thunderbird is a museum piece that you have to volunteer to maintain.

The market is full of clients that are good at one mailbox, bad at the way I actually live across four, and completely silent on the idea that a mailbox could belong to more than one person.

What a shared mailbox actually means

The thing I needed and could not find anywhere was a workspace.

Not a feature. A first class concept. I wanted to log in once, see my personal mailboxes, see my AstraQ mailboxes, and have the AstraQ contact@ box visible to anyone on the team who was supposed to be reading it. I wanted us to triage that mailbox together, with the read state and labels visible to all of us, so nobody re-reads a thread and nobody misses one. I wanted a way to invite a teammate in two clicks, scope them to one workspace, and revoke them the same way.

This was the thing the rest of the market had decided was not worth building. So we built it.

Metis has workspaces. You sign up, you get a personal workspace by default. The day you start a company, you create another workspace and connect that company's mailboxes to it. Admins can mark any connected mailbox as shared, and now every member of the workspace can open it from their own sidebar with their own permissions intact. The active workspace lives in a cookie, the active mailbox lives in a cookie, switching between them is one click. No forwarding, no Google Groups, no shared password. Just the inbox, with the right people in it.

AI in email is a different fight

This year, every email client added an AI assistant. Most of them did the same thing. The assistant lives outside the app. It takes your thread, sends it to a model on someone else's infrastructure, and returns a draft.

I cannot use any of those at work.

Half of the email I write at Deloitte cannot be passed to a third party model under any sane interpretation of the contract. Half of the email I write at AstraQ involves customer documents, technical roadmaps, or partner conversations that we are under no obligation to leak to a model provider. The honest answer for a security company is that the AI in your email cannot be a data pipeline to OpenAI or Anthropic by default. It has to be something you control.

The bolt on assistants also miss the basic point. The thing I want from email AI is not a writing helper. I can write. The thing I want is a daily triage that has read every thread I missed overnight and tells me what actually needs my attention. None of the current assistants do this well, because they were not built into the inbox. They were sprinkled on top.

Metis has an agent that lives inside the app. It has tools. It can list emails in any folder, search across them, open a thread and summarise it, star or unstar, run a phishing check, and delete with a two step confirmation so it never deletes without you saying yes. It groups your morning by Action Required, Work, Transactions, Newsletters, and Personal, the way a chief of staff would if you had one. The default model is Claude Sonnet 4.5 through OpenRouter, and the model is configurable, so you can route it through your own deployment if you have to.

There is a separate, smaller model that runs every time you open an email and writes three short reply variants for you. Accept. Decline. Need more info. Two or three sentences each. You click one and it lands in the compose window. That is the thing I missed the most about not building this earlier.

Phishing is the part everyone forgets

Phishing is the most common entry point for a real cyber incident in 2026, and email clients are still treating it like 2008.

The standard model is that the email provider runs a spam filter on the way in, and the client trusts the result. That worked when phishing emails had broken English and bad logos. It does not work now. Phishing email today is written by a model. The grammar is fine. The logos are right. The sender domain is one character off from a real one, and your eye does not catch the difference at six in the morning on a phone. The provider filter catches some of it. A lot still gets through.

The fix is not a better filter at the provider. The fix is a layer on the client that looks at every incoming email, fast, and tells you the risk before you click. Sender reputation, link inspection, language patterns, attachment behaviour. Done where the attacker cannot reach.

Metis runs a phishing analyser on every email you open. It returns a classification, Legitimate, Suspicious, or Phishing, with a risk score from zero to a hundred, a confidence number, a recommended action, and a list of reasons. The badge sits in the header of the email, colour coded. You click it to expand. The same check is exposed as a tool to the agent, so you can also ask "is this phishing?" in chat and get the same answer with the same reasons. The detection itself runs in a separate service, which means it can be upgraded independently without touching the mail client.

There is no setting to turn this off. If you have to remember to put the seatbelt on, you will eventually forget.

The boring features matter too

Most of email is the boring features. We did not skip them.

Folders. Threads. Infinite scroll. Native search across each provider. Compose with a real rich text editor. Attachments. Star, unstar, mark read, mark unread, mute a thread you no longer care about. Archive and restore. Snooze with presets and a custom time picker, and a scheduler that actually wakes the email back into the inbox at the time you chose. Schedule a send for later. Undo a send for five seconds after you click. Auto-saved drafts every 1.5 seconds, in the provider's own draft system, so you can pick them up anywhere.

Rules, the way Gmail filters work but better. Conditions across from, to, subject, body, has attachment, combined with AND or OR. Actions to archive, delete, mark read, star, add a label, or mute. Auto applied on every inbox fetch, applied retroactively with one button.

Labels. Six categories assigned automatically by a small classifier, plus custom labels with a colour palette. Click a label in the sidebar to filter the whole inbox by it. The classifier never overrides a label you set yourself.

A command palette on Cmd-K. Keyboard shortcuts for everything. Two themes, an Obsidian style warm dark and a light one. Two factor required on every account, because if you are running an email client for security people, you cannot skip that step.

The licence question, because somebody will ask

Metis Mail is source available under the Business Source License 1.1. The whole codebase is on GitHub. You can read it, fork it, run it yourself, modify it for your own use. The only thing the licence stops is someone packaging it up and selling it as a competing hosted service to ours. Every published version converts to a standard MIT licence four years after it ships.

We did not want a security product where the user has to take our word for the code. We also did not want to build for two years and watch a larger company resell our work the week we launch. BUSL plus the four year MIT clock is the honest middle.

Why we are doing this

There is a version of this story where we just used Superhuman across four accounts and got on with the day. That is the rational answer. The reason we did not is that email is the one piece of infrastructure every business in the world has agreed to use, and almost nobody is building for the way the work actually happens. Not the single mailbox of the venture backed founder. The four mailboxes of the consultant, the student, the founder who still has a day job, the partner working across two companies, the team that shares a contact@ address.

If we get this right, an analyst in a hospital, a salesperson in a bank, a freelancer with five clients, a two person company sharing a hello@ box, all of them get a calmer inbox that does not leak their data, does not miss the phishing email, and does not require them to learn three different keyboard shortcuts to get through the morning.

Metis Mail is live at trymetis.email. The code is on GitHub. If you want to talk, write to me at contact@rajpundkar.com, or book a call. The slow way is the only way.