Security & trust
The safest claim in email is the one you can check. Here's what inxo does and how you can see it.
Scoped access, proven.
Mail access is scoped to the mailboxes you choose: inxo requests no organisation-wide mail permission. Before a connection activates, a verification step shows you the evidence — which mailbox, and which permissions. Disconnecting stops the sync, erases the mail and drafts we hold for that mailbox, and records exactly what was removed. The Exchange permissions stay until your Microsoft administrator removes them: inxo lists the per-mailbox grants it created, and can remove one with their sign-in.
What we ask your Microsoft administrator for.
inxo is one multi-tenant application, and it requests no permission to read mail across your organisation. To read and answer a mailbox, setup creates access restricted to that one mailbox, plus two read-only Exchange roles, organisation-wide, that inxo uses to check that access. Documents need a separate grant per SharePoint site or Dynamics 365 environment. Before a mailbox is switched on, and every day after, inxo checks that its application holds nothing it did not ask for, and stops that mailbox if it does.Every permission, and what inxo does with it →
Approval by default.
No reply leaves without an authorized approval. Each approval is recorded — who approved, which draft version, and when. Agent auto-send is designed as an explicit, audited, per-workspace policy — never a default. Coming
approval 01a0867d-1ef0-7d98-a873-db16db2ce49f approver 01a08682-c440-7c1a-9daa-668075d7c7d2 (reviewer, Brand A) draft 01a0867c-8e68-718d-8045-320ae2141d63 v1 approved_at 2026-09-02T14:07:31Z provenance human_console auto_send: false sources ▸ pricing-2026.pdf ▸ billing-faq.md sent →| provider_message_id AAMkAD… reconciled ✓
Drafting under hard rules.
Replies stay transactional by default, and whether a draft may mention your own products is a setting you control. Personal data beyond what the writer provided is blocked on every send — that one is a wall. Promised fixes, timelines and promotional phrasing are steered and measured, with your reviewer as the control — that is true whichever way that setting is turned, because it is an instruction the model follows rather than a check that stops a reply. We would rather tell you which is which than let you assume the wall is everywhere. Bounces and auto-replies don't generate drafts. When neither the conversation nor your knowledge base contains the answer, inxo is instructed not to invent one, and the sources on the draft show you what it did use.
Inbound mail and your documents are data, not instructions.
Inbound mail — including text addressed to the AI — and the documents in your knowledge base are both treated as material for drafting a reply, never as commands. A document cannot change how inxo behaves. And underneath it, the rule that doesn't depend on the model being right: nothing sends without approval.
Your knowledge base is yours.
Each knowledge base belongs to one workspace. It is never used to train models, never shared across accounts, and deleted with the last mailbox connection for that controller.
AI-drafted text: we don’t indemnify it, and here is why.
Every message sent through inxo is yours: you are the author and the sender of record, and the moment a draft is generated it is your content, not ours. Because authorship sits with you, we do not offer to defend you against a claim that a specific piece of AI-drafted text infringes someone’s copyright, trademark or other right. We have made that choice deliberately, and we would rather write it down here than let you discover it in a clause.
What we do defend. If the inxo software itself — the platform, not the words it drafted for you — is found to infringe a U.S. patent, copyright or trademark, we defend that claim and cover the damages finally awarded. That is the ordinary software non-infringement promise, and it comes with the ordinary conditions: it covers the Service used as we provide it and in accordance with the Terms, and not a claim arising from your own content, from a change to the Service you made or directed, from your breach of the Terms, or from combining the Service with something we did not supply. §6 of the Terms is the operative text and states the exceptions and the remedy in full; this page summarises it rather than adding to it.
What holds regardless. We never train any model on your mail or your drafts, and we will not send your mail to a model provider until that provider’s no-training commitment is in force for the exact model and region in use. Not indemnifying the words changes nothing about who owns them or controls them: that stays with you throughout.
If protection against AI-output IP claims matters to your decision, ask us — and ask anyone else you are evaluating — exactly what such a commitment covers: the platform, or the generated text; which rights; and what is carved out. We would rather answer that plainly than have you assume it either way.
Your data.
inxo acts as a processor for the mail it handles. Draft retention is bounded; mail and drafts are deleted when a mailbox is disconnected. Audit records — drafts, approvals, sends — are retained in tamper-evident form. AI-generated content is marked as such.
When something goes wrong.
Incidents are severity-classified and handled through a defined process. Where your customers' mail may be affected we notify you without undue delay, because you are the controller and the regulatory clock is yours to run — we give you what you need to run it.