Email infrastructure for AI agents
The governed mailbox layer for agents.
Unlimited inboxes, your domains and your delivery, search that understands, and a real inbox for the humans in the loop. Every feature on this site carries its status, checked against the engine on every build. Beta
71 capabilities built into the engine and 117 MCP tools, over REST, MCP, JMAP, IMAP and SMTP.
# 1. The agent signs itself up. Until its human sponsor
# verifies, it may only reply.
POST /v1/agent/signup
{ "name": "research agent", "sponsor_email": "owner@example.com" }
# 2. A disposable inbox for one sign-up flow.
POST /v1/inboxes
{ "ttl_seconds": 3600, "autonomy": "send_with_approval" }
# 3. Block until the verification code arrives.
GET /v1/inboxes/{inbox_id}/wait?has_otp=true&timeout=60
# 4. Record where the work stands, on the thread itself.
PATCH /v1/threads/{thread_id}/metadata
{ "metadata": { "stage": "verified", "owner_agent": "signup-bot" } } Four promises, each with its status
-
Unlimited inboxes
An inbox per agent, per customer, per browser session or per test. Paid plans do not charge by the inbox.
- Unlimited inboxes Shipped
- Disposable inboxes with a time to live Shipped
- Pods Shipped
-
Your domains and your delivery
Send and receive on your own domains, with DKIM, SPF and DMARC records published at your DNS host or checked live, per-tenant reputation isolation, automatic suppression, and inbound mail authenticated and filtered before it is accepted. Sending through your own Amazon SES account is being verified.
- Custom domains Shipped
- DNS setup made easy Shipped
- Managed sending with per-tenant reputation isolation Shipped
- Delivery events and automatic suppression Shipped
- Inbound mail server that checks mail at the door Shipped
- Bring your own Amazon SES account Beta
-
Search that understands
Full-text and semantic search across every inbox a key can see, typed metadata and decisions on every message and thread, and queries that find threads by the state of the work.
- Search that understands Shipped
- Queries across threads by metadata Shipped
- Typed metadata on messages, threads and inboxes Shipped
- Decisions on every message Shipped
-
A real inbox for the humans in the loop
A webmail and calendar on the same mailboxes, with the approval queue, an activity timeline and a lock that lets a person take a thread over. Held sends can also be approved from Slack, a signed email link, the API or MCP.
- Webmail for the humans in the loop Shipped
- Human approval of held sends Shipped
- Approvals in Slack and by email link Shipped
- Take a thread over from the agent Shipped
- Shipped
- Built and tested in the engine today. Hosted access opens through the waitlist.
- Beta
- Built, with known gaps; interfaces may change.
- Not offered
- Not offered and not on the roadmap.
Governance
Leaving out the send tool is not a policy
An agent holding a credential that can send can send, whatever its prompt says. nofeb puts a sending policy on each inbox and enforces it on the server, for every surface. There are six tiers, from read-only to autonomous, with allowlists, per-agent caps and an append-only audit log beside them; a held send waits for a person, and an agent that signed itself up can only reply until its human sponsor verifies.
How the tiers work| Tier | What it does |
|---|---|
read_only | Sending is refused. |
draft_only | Sending is refused; the agent writes drafts that a person sends. |
send_to_allowlist | Sends to allowlisted recipients go out; any other recipient is held for approval or refused, as the inbox's policy says. |
send_with_approval | Every send is held until a human approves or rejects it. |
send_when_confident | Each send is checked by outbound safety questions and goes out when every answer clears the inbox's threshold; otherwise it is held for approval with the failing questions as reasons. |
autonomous | Sends are queued immediately, subject to sponsor gating. |
Safety
Inbound mail is untrusted input
Anyone can email an agent. Every message is scored for prompt injection, quarantined above your threshold, labelled with how far its sender can be trusted, and given a safe_body: only the new text, with hidden content and forgeable quoted history removed, inside a provenance header that tells the model whose words these are.
How inbound mail is made safeThreads as state
The thread is the agent's working memory
Keep stage, owner, follow-up time and deal value on the thread itself as typed metadata, declared once and validated on every write, find every thread at a given stage across inboxes in one query, and hand a model the whole thread in one call.
Threads as stateBuilt into the engine
What an agent can use on day one
-
Agent self-signup with a human sponsor
An agent creates its organization, first inbox and API key in one unauthenticated call. Until the human sponsor it names verifies by email, it may only reply to people who wrote to it first.
-
Unlimited inboxes
Create as many inboxes as the work needs, with a generated or chosen address and an idempotent client reference.
-
Pods
Sub-tenants inside one organization: one per end customer, created idempotently from your own customer id.
-
Scoped API keys
Keys bound to the organization, one pod or one inbox, holding only the scopes they are given. A key cannot mint a key wider than itself.
-
Wait for an email
Long-poll an inbox for up to a minute until a message matching sender, subject or has-a-code arrives.
-
One-time code and link extraction
Verification codes and classified links are pulled out of every message and returned as fields.
-
safe_body for untrusted mail
The message's new text with zero-width and direction-altering characters, hidden HTML and forgeable quoted history removed, wrapped in a provenance header that tells the model whose words these are.
-
Injection scoring and quarantine
Every inbound message gets a prompt-injection score from 0 to 1 with its reasons: hidden text, invisible and look-alike characters, instructions aimed at an AI, role and chat-template tokens, tracking beacons, data-carrying links and encoded payloads. Above the threshold you set per organization, pod or inbox, the message is quarantined (or only flagged) for a person to release or delete, and its content stays out of the delivery event.
-
Send, reply, reply-all and forward
Threading headers are set for you, replies default to the original sender, and forwards carry the original attachments.
-
Autonomy tiers enforced by the server
Every inbox carries a sending policy, from read-only to autonomous, enforced on the server for every surface that sends, so switching from the API to MCP cannot bypass it.
-
Human approval of held sends
A held send waits with status awaiting_approval until a human approves or rejects it; both steps emit events. The approval queue lists every held send with its recipients (each marked if new to the conversation), the reasons it was held and who submitted it. Approving needs the separate send:approve scope, which keys from agent self-signup never hold, so an agent cannot approve its own send.
-
Recipient and sender allowlists and blocklists
Lists of addresses, domains or wildcards per organization, pod or inbox: who an agent may write to, and whose mail it reads. A blocked recipient is refused on every autonomy tier, a blocked sender's mail is rejected or quarantined, and you choose whether a send outside the allowlist is held for approval or refused. A check call predicts what a send or an inbound message would meet before it happens.
-
Typed metadata on messages, threads and inboxes
Merge-patch keys onto a message, a thread or an inbox, and declare a key's type so every write to it is validated.
-
Follow-up timers
Set follow_up_at on a thread and, when that time passes, a thread.due event carrying the thread's metadata goes to your webhooks and the event feed, so agents need no scheduler of their own.
-
Search that understands
Full-text and semantic search over subjects, addresses, the new text of each message, attachment names and attachment text, across every inbox a key can see. Filter by inbox, pod, folder, thread, sender, recipient, date, attachments, keywords, authentication failures and metadata; results carry highlighted snippets and can be grouped by thread.
-
Decisions on every message
A default set of typed decisions on every inbound email (intent, urgency, needs-reply, automated, verification email, phishing, injection risk, requires-human and language) with a probability each, your own decision schemas beside them, human relabels that correct and retrain it, per-question accuracy and calibration stats, and confidence-gated sending for the send_when_confident tier. Decisions run without leaving nofeb by default; a third-party model sees mail only if you opt in.
-
Signed webhooks with required scope
Standard Webhooks signatures, retries and a delivery log. Scope (organization, pod or inbox) is required, so a subscription never silently widens to everything.
-
MCP server
A remote MCP server over streamable HTTP whose tools are generated from the same operation table as the REST API and dispatched through the same router, so a tool cannot behave differently from its endpoint.
-
MCP sign-in with OAuth
MCP clients connect by signing in instead of pasting a key: OAuth 2.1 with discovery, dynamic client registration, client ID metadata documents, PKCE, tokens bound to the MCP server, refresh-token rotation and immediate revocation. A person approves the scopes, and optionally one inbox, in the console, and can disconnect the app at any time. API keys keep working beside it.
-
Calendar per inbox
A calendar on every inbox: events with recurrence and time zones, invitations, updates, cancellations and replies by email, auto-accept rules, free/busy, availability and booking that sends the invitations, MCP tools for scheduling, and CalDAV and JMAP Calendars for standard clients. Invitations an agent sends follow its inbox's autonomy tier.
-
IMAP and POP3
IMAP4rev2 (with IMAP4rev1 compatibility, IDLE, CONDSTORE and QRESYNC) and POP3 on every inbox, over TLS only, signed in with the inbox address and an API key or a revocable per-inbox app password.
-
Managed outbound delivery
nofeb delivers your outbound mail from your verified domains, signed with DKIM and aligned for SPF and DMARC. There are no mail servers or sending IPs for you to run.
-
Inbound mail server that checks mail at the door
nofeb's own MX receives mail for the shared domain and your custom domains, with STARTTLS. It checks every recipient before accepting any data, verifies SPF, DKIM, ARC and DMARC, and scans for spam and viruses before it answers. Mail that fails a DMARC reject policy, carries a virus or is clearly spam is refused during delivery instead of bounced later; likely spam and DMARC-quarantined mail is filed into Junk. Every message carries a security verdict with its authentication, spam, virus and TLS results.
-
DNS setup made easy
nofeb detects who hosts your domain's DNS and gives you the shortest path for that host. It can publish the records through the Cloudflare, DigitalOcean, Hetzner DNS, Vercel, Porkbun or Gandi API with a token used for that one call and never stored, and it never deletes a record. Every other host gets step-by-step instructions in its own terms, plus zone-file, Route 53 and Terraform exports. A live check shows each record's status, what was found and the likely mistake. Pointing your MX at nofeb is a separate choice, so publishing sending records never moves your mail.
Who it is for
Built for the teams that ship agents
- Agent developers
- QA and browser automation
- AI SDR and outbound
- Support automation
- SaaS platforms
- Finance and AP
- Operations (logistics, real estate)
- Assistant builders
- MCP users
- Platform and security leads
Start with a guide: Give an AI agent its own email address · Receive verification codes in Playwright tests · Test email verification in Cypress · Add email to Claude Code, Cursor and other MCP clients.
Questions
nofeb in three answers
What is nofeb?
Email infrastructure for AI agents: inboxes an agent can create for itself, mail scored for prompt injection and made safe for a model, typed state and decisions on every thread, search, a calendar per inbox, and a sending policy the server enforces, over REST, TypeScript and Python SDKs, a remote MCP server, JMAP, IMAP and SMTP.
How is nofeb different from a sending API?
A sending API sends from an application. nofeb gives each agent an inbox: it receives, threads and stores mail, waits for codes, and limits what the agent may send with autonomy tiers, allowlists, caps and human approval, with a webmail where people share the mailbox.
How does mail reach a nofeb inbox?
Through nofeb's own MX, which authenticates senders and checks for spam and viruses before accepting, or through Cloudflare Email Routing, Postmark, Mailgun, SendGrid or any raw POST.
Solutions
What agents do with email
- Browser agents and sign-upsAn inbox for every browser session
- AI SDROutbound and reply handling for AI SDRs
- Support triageSupport agents that know when to hand over
- Finance and APInvoice intake for AP agents
- Logistics and freightQuote threads for freight agents
- Real-estate closingDocument chasing for closings
- SaaS platforms with podsAn inbox for every one of your customers
- Personal assistantsEmail for personal and executive assistants
- QA of email flowsThrowaway inboxes for testing email flows
Pricing
Inboxes are not the unit
Start on Free with 10 inboxes and 3,000 messages a month. Build is $19 a month with unlimited inboxes, and usage past the allowance is metered, not blocked.
Plans and unit pricesCompare
Sourced, dated comparisons
What each alternative does, from its own documentation, and when it is the better choice.