AI drafts every reply.
You keep the send button.
A complete business mailbox with an agent that reads, triages and drafts — then stops. Every reply waits for your approval. Self-hosted, runs on local AI, built to resist prompt injection.
Re: SSL certificate order #48213 failed
Provisioning returned an error. Can you confirm the domain and resend the CSR?
A full business mailbox, with a supervised AI inside it
Professional email with an agent that does the reading and the first draft, then hands the decision back to you.
Classify, draft, self-check, score. Then wait.
Every incoming message runs the same pipeline: work out what it wants, pull in the thread and your knowledge base, draft a reply, check it, and score how confident it is. Then it stops.
- Drafts grounded in the thread and your knowledge base
- Confidence is a signal for you, never a licence to send
- Every incoming email is treated as untrusted input
Important mail on top, and rules that run themselves
AI-ranked Important view, automation rules per mailbox or across all of them, fast full-text search and proper threading — for every account you connect.
- Several mailboxes behind one sign-in, or one unified inbox
- Search every folder, with match context in the results
- Unsafe or suspicious mail kept out of Important
Ask about anything, including the screenshot
The assistant reads the open thread, its attachments and inline images. Paste an error screenshot, ask what it means, and get an answer grounded in the conversation.
- Reads inline screenshots, images and documents
- Suggests actions; you confirm every one
- Images and attachments are untrusted too — guardrails hold
Start the day with a briefing, not a backlog
One screen that tells you what matters: what came in, what’s important, who is still waiting on you, and which AI drafts are ready for review — with today’s AI usage against your budget.
- Important unread, ranked by AI importance
- Awaiting reply: threads you haven’t answered in 12 hours to 30 days
- The draft queue, with confidence per draft
- Optionally open on the Briefing every time you sign in
The feeds you follow, triaged by the same AI
Follow RSS and Atom feeds or import an OPML file. Every article gets an importance score, a severity and a summary, with CVE IDs pulled out, so the one story that matters doesn’t drown in fifty that don’t.
- Security, tech and general categories, with topics you can mute
- Summaries and translation to English, Bulgarian or German
- A daily or weekly digest in your inbox, filtered by severity
- Ask the assistant about any article
Built for people who don’t hand AI the send button
The drafting agent can’t send, forward or delete; it only returns text for you to review. Every incoming email is treated as attacker-controlled input and defended against prompt injection. Your mail, keys and models stay on your infrastructure.
Nothing leaves your outbox without a human decision.
Email content is data, never instructions. Attempts are flagged.
Mailbox credentials and settings are encrypted with a key that never lives in the database.
Role-based access, TOTP two-factor sign-in, and a full audit trail.
A hard cap on AI usage, so a burst of mail can’t run up your bill.
Encrypt, sign and verify. AI stays off encrypted threads unless you opt in.
Three steps. You control every one.
Connect your mailboxes
Any IMAP/SMTP account. Credentials are encrypted; your sign-in is a separate identity with its own 2FA.
The agent drafts
It reads the thread, pulls in context, writes a grounded reply, checks it and scores its confidence.
You decide
Approve, edit or discard. Nothing is sent until you say so.
Review drafts with one thumb
The same supervised drafts, the same theme, on a single-column layout with a bottom tab bar for Mail, Briefing and News. Add it to your home screen and it runs like an app.
- Approve, edit or discard with full-width controls
- Installable as an app from the browser
- Your theme follows you across devices’ browsers
Invites handled where they arrive: your inbox
Meeting requests already come in as email. The calendar turns them into events with one-click replies, syncs CalDAV and Microsoft 365 / Teams calendars, and reminds you before a meeting starts. The AI can suggest times; it never books behind your back.
Accept, tentative or decline, with conflict warnings.
Two-way with Nextcloud, Fastmail, iCloud and others.
Microsoft 365 calendars and Teams meeting links.
In the app, as push notifications, or by email.
Straight answers
Does the AI ever send email on its own?
No. The agent only drafts; every reply waits for a person to approve it. Any future autonomy will be opt-in and limited to senders you explicitly allow.
Where do my mail and the AI processing live?
On your own infrastructure. Each deployment is a separate, self-hosted instance and can run entirely on local AI models, so mail content never has to leave your network.
Which mail providers work?
Any mailbox that speaks IMAP and SMTP: Gmail (with an app password), Fastmail, iCloud, or your company’s own server. Microsoft 365 and Teams integration is being built together with the calendar.
Can someone hijack the AI with a crafted email?
That is the attack the design starts from. Every incoming body is wrapped as untrusted data the model must never obey, injection attempts are flagged, and the drafting agent has no ability to send, forward or delete in the first place.
Can the look match our brand?
The page and the app share the same ten themes, and each person’s choice is remembered. Per-instance logo, colours and copy are on the roadmap.
Your inbox, with a co-pilot that asks first
Sign in to your mailbox. Your drafts are waiting for review.