Skip to content
nofeb

Solutions / Support triage

Support agents that know when to hand over

For support teams putting agents on the front line

Support mail is the most public inbox a company has. Anyone can write to it, and the agent that answers it must know what it can resolve and what a person should see.

The problem

A support agent reads mail written by strangers, some of it written to manipulate it. It needs the thread in context, a place to record the ticket's state, and a hard limit on what it may send without a person.

How it runs on nofeb

  1. Receive mail on a support inbox and subscribe a webhook scoped to it.
  2. Every message is scored for prompt injection on arrival. Above the inbox's threshold it goes to quarantine, where a person releases or deletes it; the agent never sees its content, and message.received does not carry it.
  3. Decisions on urgency, intent, needs-reply and requires-human arrive on each message. Rules act on them before the agent does: label urgent mail, move automated notices out of the way, or forward billing disputes to the finance inbox.
  4. Before answering, the agent searches past tickets across every support inbox by meaning as well as by words, filtered by sender domain, date or ticket metadata, and reads the closest threads.
  5. Give the model safe_body and the thread context, not raw HTML, and record ticket id, category, priority and SLA on the thread as typed metadata.
  6. Run new categories at draft_only, where the agent writes a draft and a person edits and sends it, or at send_with_approval, where every reply is held until someone approves it. Raise the tier once you trust a category.
  7. People decide in the webmail's approval queue, which shows each held reply with its thread, its recipients and why it was held, or from Slack, where the request arrives with Approve and Reject buttons.
  8. When a ticket needs a person, they take the thread over. Until they hand it back, the agent's sends on that thread are held or refused, so it cannot answer over their shoulder.

Where nofeb stops

  • nofeb is not a help desk. It has no customer portal, macros or reporting; pair it with the help desk you use, or build on the API.

Status of everything this relies on

CapabilityWhat it doesStatus
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. Shipped
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. Shipped
LLM-ready thread context A whole thread, oldest first, with safe_body text and the thread's metadata, in one call. Shipped
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. Shipped
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. Shipped
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. Shipped
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. Shipped
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. Shipped
Rules and Sieve Declarative rules over sender, recipients, subject, body, headers, attachments, size, the security verdict, metadata and decisions, that label, move, flag, set metadata, quarantine, forward, auto-reply or reject. Test rules against a stored message with a per-rule trace, and import or export Sieve, with anything that cannot be mapped reported line by line. Shipped
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. Shipped
Drafts Agents write drafts that a person edits and sends, over the API, MCP, JMAP, IMAP and the webmail. It is the documented path for draft-only inboxes: sending a draft needs the approval permission agents do not hold. Shipped
Take a thread over from the agent A person takes a thread over; until they hand it back, the agent's sends on that thread are held for approval or refused, and an event records who is driving. Works over the API, MCP, JMAP and the webmail. Shipped
Webmail for the humans in the loop A full webmail and calendar on the same mailboxes the agents use, with an agent-activity timeline, the approval queue, a per-thread driver lock and a metadata panel, signed in through the same accounts as the console. It opens with hosted access. Shipped
Approvals in Slack and by email link Approve or reject a held send from a Slack message with Approve and Reject buttons, or from a signed, single-use, expiring link in an email. Slack requests are verified with your app's signing secret and can be limited to named people; the email link decides only on an explicit click, so link scanners cannot approve. Shipped