What runs today
Envolt Mail is in development. This page is generated from the same status file as every badge on this site, so the two can't disagree.
- Live
- Public and running in production.
- Running in staging
- Built and running on our staging environment, on synthetic test mail only.
- Designed
- Specified in detail, not built or not switched on yet.
Capabilities
Public
- This websiteLive
envoltmail.com, with a strict content security policy, no cookies and no analytics.
- Authenticated, signed domainsLive
Our own domains publish Sender Policy Framework (SPF), DomainKeys Identified Mail (DKIM) and Domain-based Message Authentication, Reporting and Conformance (DMARC) records and are signed with Domain Name System Security Extensions (DNSSEC).
Inbound
- Agent mailboxes on our own domainsLive
Each of 25 of our own domains has an agent address on its own agents subdomain, separate from the people's mail on the main domain, which stays where it is. Mail to it is screened and counted without storing its content, and nothing is answered yet. An operator-run delivery check sends one labelled message from every domain to the next and confirms it arrived with SPF, DKIM and DMARC passing. The DNS records behind each agent subdomain are re-checked every day without sending anything; a failed lookup is reported as unknown, never as healthy. The Agent Seal signature does not yet survive our current sending provider route.
- Airlock quarantineRunning in staging
Parses each message in isolation, removes supported hidden-text patterns and invisible characters, and flags comments, encoded blocks and unresolved styling. Detectors also flag summary manipulation, look-alike senders and forged quoted replies.
- Rules-first triageRunning in staging
Duplicates, bulk mail and notifications are recognized by rules before any model runs.
- Agent Seal verificationRunning in staging
Incoming mail that carries an Agent Seal header is checked against the sender's published keys.
- Quarantined readerRunning in staging
A reader with no tools describes each message. Suspicious input or output gets a fixed review notice and no extracted actions. Sensitive numbers are redacted. It runs on a no-cost synthetic stand-in; no real model is switched on.
Storage
- Per-message encryptionRunning in staging
Raw mail and attachments are encrypted with a key per message, wrapped under a per-customer key, with versioned keys.
Record
- Tamper-evident ledgerRunning in staging
Every event is hash-chained and anchored daily. Conversation threads are re-verified before each anchor, and a broken chain is sealed and recorded, never anchored.
- Isolation canariesRunning in staging
Every staging check plants unique markers in separate synthetic test customers and confirms neither can see the other's.
Mailboxes
- Mailbox connectionsDesigned
Read-only connections beside your existing mail, never in its delivery path. A read-only connector is built and tested but not connected to any mailbox. The pilot built for our own mailbox keeps counts and, for one day, the minute and label of each message; never a sender, subject or text.
Agents
- Agent draftingDesigned
Agents read a conversation and propose a plan: a reply, an action, or a question for a person.
- Persona roomsDesigned
A studio, a pen name, a stage name and each project are separate rooms with their own agents, voice and contacts. Nothing crosses between rooms unless you link them.
- Seal policyRunning in staging
A policy engine that checks every plan's recipients, content, rate and approval needs against rules your team owns. In staging it judges synthetic plans; it gives the same verdict as its written specification on every published test case.
People
- Console previewRunning in staging
A morning brief and decision cards for the accountable person, built from the synthetic ledger. It also lists what agents were stopped from sending, and why.
- Approval queueRunning in staging
One card per decision: the ask, a recommendation, alternatives, the risk, and approve, edit, hold or delegate. In staging, an approved reply waits in a send-hold with Undo, is re-checked when the hold ends, and only then gets its single-use capability, into the sink. Drafts come from test scripts, since agents do not draft yet, and the signed-in console does not open this queue yet.
- Earned autonomyRunning in staging
An agent can earn routine sends from its measured track record, and loses them automatically on any serious mistake. In staging the Autonomy Ledger projects evidence from real decisions on synthetic mail; promotion is always a card for a person, never automatic. No agent has earned anything.
- Kill switchRunning in staging
Halts sending for a whole customer, one persona or a single agent. Engaging it cancels every outstanding capability at once, and releasing it is a recorded human action. A halt across all customers at once exists too. Runs in staging on synthetic test customers.
- Passkey sign-inRunning in staging
Sign-in for the accountable person uses passkeys only: no password and no email link. Anything that makes the system less safe, such as releasing a halt or starting a mailbox pilot, needs a fresh passkey check; making it safer never does. After signing in, that person sees a console: a setup checklist, the 90-day clock, the day's totals, label review message by message, a daily label check, the pilot controls and the kill switch. A send whose outcome was lost is shown at the top for a person to check.
- No standing staff accessDesigned
Staff have no standing access to anyone's mail content. Support access happens only with the customer's approval, time-boxed and recorded where the customer can see it. No customer mail exists yet, so the approval flow is not built.
Outbound
- Labeled, sealed sendingRunning in staging
Each allowed message gets a single-use capability that is spent before anything is sent, and is rendered with the AI agent label, a signed Agent Seal header, a plain-language disclosure and a link to its receipt. The hand-off to the mail provider is written to the ledger first, so a message whose outcome was lost is held for a person and never sent twice. In staging the message goes to a sink that sends nothing. In production the only messages sent are delivery checks between our own agent mailboxes, through Brevo.
- Public receiptsRunning in staging
Each send gets a signed receipt that shows who the agent is, who it works for and whether a person approved, never the message content. In staging receipts are issued and signed, and the link in the message opens a public page that anyone can read. The receipt is prepared and signed before the message leaves. Besides made-up staging mail, the only messages with receipts are the production delivery checks between our own agent mailboxes.
Builders
- API, webhooks and MCPRunning in staging
An agent API with scoped keys runs in staging, over REST and as a Model Context Protocol (MCP) server: list conversations page by page, read their record, see what was stopped, see the decision cards raised for a person, read the brief, propose a reply. An agent can read decisions and cannot approve them. Signed webhooks are specified, not built.
Mail pipeline
- Evidence and risk-bound actionsRunning in staging
Atomic thread risk, revocation on changed classifications, source quote references, bounded isolated text attachments and a reusable intake component. The Lab previews inert review and publishes fixture misses and review burden. No real mailbox or paid model is enabled.
Evidence
Release receipts, daily ledger anchors, the 90-day clock and the list of what is not proven yet are on the proof page.
What comes next
The planned order and the eight gates that must pass before outside teams join are on the beta page.
History: the changelog (also as an Atom feed). Agents: this page is at /status.md.