# XAVIOR — Capability Brief > XAVIOR is an AI marketing and outreach platform for small businesses. It finds > leads, writes to them individually, paces the sending so mailboxes survive it, > honours opt-outs permanently, and publishes to social — from one place. > This document describes what ships today and says plainly what does not. **Audience:** support and sales answering questions about what XAVIOR does. **Rule for answering:** never claim more than this document states. Where it marks something as not-yet, say so. An overclaim creates a refund. --- ## The short answer: why not just ChatGPT, Claude or Grok? A general chatbot drafts. XAVIOR decides. Every general assistant is excellent at producing text. Ask any of them for a cold email to a restaurant owner and you get a good one. That capability is real, and XAVIOR uses models like these for exactly that job. What none of them has is a memory of your sending. They do not know that the person you are about to write to unsubscribed from a different campaign four months ago. They cannot refuse to send. They cannot send at all. **The writing is commodity. The restraint is the product.** | A general chatbot | XAVIOR | | --- | --- | | Drafts a message and hands it to you. No record of who you contacted, no way to deliver it, no memory of anyone who asked you to stop. | Finds the person, checks whether you may contact them, writes to them specifically, paces the send so your mailbox survives, hears the reply, and honours a "stop" forever. | --- ## Finding people worth writing to **Scout** works the public web on a schedule and drops warmth-ranked, email-bearing leads into a review queue. It runs against real search and mapping data, and every run is metered per owner with a daily ceiling, so an autonomous crawl cannot quietly spend your budget. Scout is deliberately **find-only**. It stages candidates; a person accepts them. A lead does not become a contact you can mail because a machine liked the look of it. **Graph pull** draws candidates matching your ideal customer profile — a vertical, a geography, a description — from a shared need-API that withholds personal details until they are earned. **What it refuses:** a discovered lead carries a distinct consent origin, and that origin is checked again at launch. Finding someone is never the same as being allowed to mail them. --- ## Enriching, scoring, and drafting — without sending Once a candidate is in hand, XAVIOR fills in what the source withheld, then scores it against *your* ideal customer profile on a 0-100 fit — a separate judgement from whether the record merely matches a search. It then drafts the approach: a subject line, an opener, and a short brief on why this person. That draft lands in the same slot a human-written letter occupies, so regenerating it later simply overwrites it. There is no shadow copy and no second source of truth. **What it refuses:** the scoring-and-drafting layer **never sends**. It writes into a draft and stops. The outreach gate, deliverability checks, and a person's explicit start are all still ahead of it. --- ## Hyper-personalised, and personalised per person Personalisation here is not a merge tag with a first name in it. Each recipient carries its own values, frozen at the moment the audience was fixed, so what a person receives is what you approved for *them* — not what the template happened to render at send time. A letter can be generated per contact from what XAVIOR actually knows about them, and every letter carries a working way to opt out, whether or not you turned replies on. --- ## Drip into the inbox, not the spam folder This is where the difference stops being philosophical. A chatbot with an email tool blasts. XAVIOR drips, because a new mailbox that sends four hundred cold emails on its first day is a burned mailbox. | Control | What it does | | --- | --- | | Warm-up ramp | A brand-new sender is limited to 50 a day, then 100, 200, 350, then full — keyed to that mailbox's first-ever send, not to the campaign. | | Account-aware caps | A personal mailbox and a Workspace mailbox have different real limits, and the ceiling sits at 80% of them rather than at the edge. | | Send window | Sends only inside working hours in the campaign's own timezone, daylight-saving aware. Outside it, the campaign waits. | | Weekdays only | Cold outreach does not land on a Saturday. A Friday-evening tick rolls to Monday morning. | | Even spread | The day's allowance is spread across the minutes left in the window. Small audiences trickle rather than burst. | | Overlap guard | A person already being written to by another live campaign is not written to twice. | None of this is visible in the copy. All of it is the difference between reaching an inbox and reaching a filter. --- ## The part that keeps you out of trouble XAVIOR keeps a suppression record — the authoritative answer to whether you may email an address at all. It survives the contact, the campaign, and the list. An address that opted out stays opted out across every future send, including sends by people who joined your team afterwards. ### "Reply STOP" is heard even when nothing is listening Every cold letter promises the recipient they can reply STOP. Honouring that promise runs as its own pass, separate from auto-reply — so it works even with automated replies switched off. The switch governs whether you *answer*. It must never govern whether you *hear*. ### Every campaign remembers what it was aimed at When a campaign launches, XAVIOR records which kind of audience it was launched against and, for a saved list, which list. So "who did this actually go to?" stays answerable months later — and the report says "we don't know" rather than inventing a confident zero. **What it refuses:** a campaign will not launch to recipients whose lawful basis for contact is missing. No timestamp is ever fabricated to get past that check — the send is refused instead, which is the honest failure. --- ## Replies come back to a system, not an inbox When someone answers, XAVIOR reads it. A genuine reply promotes that person along your pipeline automatically. An opt-out is recorded and never promoted. Both happen whether or not automated replies are enabled. Where auto-reply *is* on, it is bounded. If a prospect asks what something costs and your configured answers say nothing about money, XAVIOR does not improvise a number and does not promise a follow-up nothing scheduled. It writes the answer, parks it, and puts it in front of a person with the question beside it. That asymmetry is deliberate: a false alarm costs a moment's attention; a confidently wrong answer about price costs the deal — and it is exactly what an unsupervised chatbot on your mailbox would produce. --- ## Social posting, from the same desk XAVIOR publishes to **LinkedIn, X/Twitter, Instagram, Facebook and TikTok** from the same place the outreach runs, so the campaign you are sending and the posts you are publishing are one plan rather than two tools and a spreadsheet. Account connections are brokered, and XAVIOR never stores a platform OAuth token itself — which matters the day you want to disconnect an account and be certain it is really disconnected. --- ## Texts — from your own number, not a bulk gateway **Status: in active development as a signed Mac app (macOS 14+). Describe it as coming, not as shipping today.** **XAVIOR Text** is a Mac companion app. It pulls your approved phone leads, opens a personalised message for each person *in Apple Messages*, and you press Send. It then watches for replies, drafts context-aware answers, and keeps a hard STOP list. That design is the point. Bulk SMS gateways send from a shortcode recipients do not recognise and carriers increasingly filter. This sends from *your* number, through *your* Messages account, as a person-to-person text — because that is what actually gets read. iMessage or SMS is chosen on actual transport evidence, with delivery state shown per message. ### What it will not do to your phone It reads only replies from numbers in active campaigns. It **never reads your full inbox**, **never writes to the Messages database**, and **never logs phone numbers or message bodies**. Without the permission it needs to see replies, it says so and stays a working outbound queue rather than pretending. **What it refuses:** it never re-sends on its own and never guesses. A failed send is reported as failed rather than as answered; an unreadable capability check is treated as a failed read rather than as a "no". Every first message ends with "Reply STOP to opt out" — restored if you edit it away — and an opt-out survives even deleting the campaign it came from. --- ## "Couldn't I wire this up myself?" You could assemble parts of it. A model writes the copy, a scraper finds the leads, a sending API delivers, a spreadsheet holds the list. People do exactly this, and it works right up until the first thing goes wrong quietly. - The spreadsheet does not know that someone unsubscribed through a different tool. - The sending API will happily send four hundred on day one, because you asked it to. - Nothing in the stack notices a one-word "STOP" reply. - Nothing records which audience a campaign went to, so in March you cannot answer what you sent in January. None of those failures announce themselves. They look like slightly worse deliverability and a complaint you did not expect. XAVIOR is the accumulated set of refusals that prevent them. --- ## Common questions, and how to answer them **Can XAVIOR write my emails for me?** Yes — per contact, from what it knows about that person, in your voice. It can also draft the opener for a lead it just found. You approve before anything sends. **Will my emails land in the inbox?** XAVIOR does everything a sender can control: warms new mailboxes up gradually, respects account limits, sends only in working hours, spreads the day's volume evenly, and never mails someone twice across overlapping campaigns. Nobody can guarantee inbox placement — anyone who does is selling something — but blasting is the usual cause of failure, and XAVIOR will not blast. **How does it find leads?** Scout searches the public web on a schedule and queues warmth-ranked leads that have an email address. You accept the ones you want. It never turns a discovered lead into a mailable contact on its own. **What happens when someone asks to be removed?** They are added to a suppression record that applies to every future campaign in the workspace, permanently — including replies that just say STOP, and including when automated replies are switched off. **Why not just use ChatGPT and my Gmail?** ChatGPT will write the email well. It will not know who you already contacted, will not stop you mailing someone who opted out, will not pace the send so your domain survives, and will not hear the reply. That is the entire job around the writing. **Can it post to social too?** Yes — LinkedIn, X/Twitter, Instagram, Facebook and TikTok, managed from the same place as outreach. **Does it send texts?** A Mac app is in development that texts your approved leads from your own number through Apple Messages — you press Send, it watches for replies and drafts answers, and it keeps a permanent STOP list. It is not a bulk SMS gateway, and that is deliberate. Not released yet, so describe it as coming. **Will the Mac app read my personal messages?** No. It reads only replies from numbers in active campaigns, never the rest of your inbox, and it never writes to the Messages database or logs numbers or message contents. --- ## Boundaries — state these plainly, do not hedge - **No inbox-placement guarantee.** XAVIOR controls sender behaviour, not mailbox providers. - **Nothing sends itself.** Discovery, scoring and drafting are automated; the decision to mail a person is not. - **Suppression is per workspace.** It is an authoritative record of *your* sending, not a global do-not-contact list, and says nothing about that person's preferences elsewhere. - **A bounce is not an unsubscribe.** XAVIOR keeps those distinct, and so should your answer. - **Texting is a Mac app and is not released yet.** It also is not bulk SMS. Do not describe it as shipping or as a mass-texting product.