Every inbound message becomes a drafted reply before your agent opens it

Most inboxes show you a list of unread messages and leave the thinking to you. Velaro's Focus Inbox does the thinking first. By the time an agent opens a conversation, it's already been triaged, clustered with related threads, checked against your knowledge base, and — in most cases — has a drafted reply sitting there waiting for a human to read, edit, and send.

This isn't one feature. It's four pieces that work together: a triaged, clustered inbox; an AI drafting pipeline grounded in your own ticket history and knowledge base; a data-scrubbing layer that strips financial and personal data out of every message before it's stored; and the option to run the whole thing on an AI key you control instead of Velaro's shared one. Here's how each piece actually works.

The inbox sorts itself before an agent ever looks at it

Every morning, Focus Inbox reads the messages that came in overnight and classifies them into buckets — what needs a reply now, what's a decision waiting on someone, and what can wait. Your team sees that briefing before opening a single message. Through the day, incoming conversations get an automatic triage label — Urgent, Needs Reply, Waiting, Offer — based on sentiment, message age, and content, so the inbox sorts by what actually needs attention instead of by arrival time.

When multiple customers write in about the same underlying issue — a shipping delay, an outage, a pricing question that's suddenly trending — Focus groups those threads into a single cluster instead of leaving five agents to work five separate copies of the same problem. Each cluster moves through a pipeline board (new, in progress, resolved) so the whole team can see where an issue stands without someone maintaining a spreadsheet.

How an inbound message becomes a drafted reply

A drafted reply is a full pass through your own data, not a generic autocomplete. When a new ticket or email comes in, Velaro pulls the conversation's message history, checks your knowledge base for grounding — running both a vector search over your indexed content and a lookup against specific KB articles — and hands all of it to the AI model along with your team's configured tone and response style. The output gets saved as a draft attached to that ticket, tagged with which knowledge base articles it drew from and a confidence score, and it sits there marked pending until an agent reviews it.

Canned response templates sit alongside this — reusable replies your team already writes by hand, organized by team and optionally linked to a specific knowledge base article, so a consistent answer stays consistent no matter who sends it. The AI draft and your canned templates pull from the same knowledge base, so an agent editing a draft is never contradicting language your team has already agreed on.

Financial and personal data gets scrubbed before it's ever stored

Message scrubbing is a redaction layer that runs on every inbound and outbound message across every channel Velaro supports — web chat, email, SMS, WhatsApp, voice transcripts, and AI-generated replies alike. It detects credit card numbers, bank account and routing numbers, SSNs, and CVVs, and strips them before the message is stored or shown to an agent. The detection is context-aware on purpose: a bank account pattern only fires near words like "account" or "routing," and a SIN pattern only fires near "social insurance," so a customer's order number doesn't get mistaken for sensitive data and redacted by accident.

For teams handling financial services conversations, that matters at the message level, not just at the payment level. A customer support agent doesn't need a card number typed into a chat window to sit in your conversation logs indefinitely — scrubbing means it never gets there in the first place, on either side of the conversation. It's a site-level setting, so it's on for the accounts that need it without changing behavior for the ones that don't.

"A customer typing 'my card number is 4111-1111-1111-1111' into a support chat shouldn't mean that number lives in your logs forever. Scrubbing catches it on the way in and on the way out."

Run the AI on your own key instead of Velaro's

Bring-your-own-key AI lets a site connect its own OpenAI, Azure OpenAI, or Anthropic API key and have Velaro's AI features — drafting, triage, clustering — run against that key instead of Velaro's shared infrastructure. The key is stored per site, write-only in the API, and every call made against it is logged to a dedicated audit trail so you can see exactly when your key was used and for what.

Switching to your own key doesn't change what the drafting pipeline does — the same ticket context, the same knowledge base grounding, the same pending-draft workflow. It changes whose infrastructure and whose contract the request runs against.

The Bottom Line

An inbox that sorts itself, drafts that write themselves against your own knowledge base, financial data that never gets stored in the first place, and the choice to run all of it on your own AI key — that's the whole pipeline, and it runs on every message the moment it arrives, not after a backlog builds up.