An AI inbox is an email client where AI is part of the operating model, not a button bolted onto compose. It can decide what needs attention, label incoming mail, draft replies, and keep memory about people and groups.

That power cuts both ways. If the AI inbox is right, it feels like a chief of staff. If it is wrong and opaque, it is just another place for work to disappear.

A useful AI inbox should expose five things

  1. Reasoning: why a message was labeled, prioritized, or drafted.
  2. Confidence: whether the model is certain or guessing.
  3. Sources: what thread, contact, or rule influenced the output.
  4. Feedback: what the user corrected last time.
  5. Control: whether the system can act or only suggest.

Postscript's answer

Postscript stores AI behavior as durable product state. Model outputs include model name, prompt version, input hash, structured JSON output, confidence, and reason. Draft feedback and label corrections are saved so future behavior can improve instead of staying stateless.

Inbox layerConventional emailPostscript direction
LabelsManual rules or opaque AINatural-language criteria plus assignment reasons
DraftsBlank composer or one-off rewriteSuggested replies with tone context and approval
ContactsAddress bookTone and relationship memory
GroupsFolders or listsDefault style guidance by relationship type
SafetyUndo after actionReview before action

FAQ

What is an AI inbox?

An AI inbox is an email client that uses AI for core inbox workflows such as triage, labels, summaries, drafts, reminders, and contact context.

Is an AI inbox safe?

It depends on controls. Prefer tools that keep sending manual by default, expose reasoning, and let users disable AI processing for sensitive senders.

Does Postscript replace Gmail or Outlook?

Postscript is designed to work with Gmail and Outlook accounts through provider integrations.