Wisp runs your techs’ day.

Wisp runs your techs’ day. Each person gets one, in any role. It watches that person’s customers continuously instead of waiting for a morning dashboard check, keeps one durable thread across all of them, reaches them on their phone with one subject and one next step, asks for approval before anything consequential, remembers what it is told, and hands bounded work to the specialist agents. It never acts outside the authority of the human it works for.

The workers do the work when you ask or when an alert fires, one job at a time, on your own model key, and they are free in every Breeze install. Wisp comes to you. It wakes on events and on a timer, keeps a running picture across your customers, and decides whether a person needs to know. The workers are hands you pick up. Wisp is already working when you arrive.

It reads what it may see, drafts ticket replies, nudges a technician, opens a ticket from an alert that a rule already says becomes a ticket, re-runs a check, schedules a sweep, and stages work in a sandbox. That is Tier 1 of the risk engine plus any Tier 2 action you have chosen to auto-approve. Everything else waits for a yes.

Act outside the authority of the human it works for, or let a specialist agent exceed that authority. Mechanically, Wisp runs under a ceiling: the technician’s own live permissions, intersected with the grant they gave it and with your AI agent policy. The ceiling is checked before every read and every action. Wisp adds no new risk tier and no bypass.

Anything consequential never ran without a yes: it waited on an approval card. What Wisp does alone is reading, drafting, nudging, re-running checks and staging work, and every one of those is in your audit log as “Wisp, for” the person. You can pause it, turn its proactivity down, and retract anything it remembered. An admin can revoke the grant at any time. And we ask founding members to tell us, in public, when it is wrong.

Preferences, standing instructions, facts about a customer, decisions, what happened after a nudge, and what was done in a working session, with the evidence. It writes memory only from what people say and from observed outcomes, never from guesses about people. A technician’s own memories are visible to them, a customer’s memories to the people with access to that customer, and every memory can be confirmed or retracted. Memory is context. It can never approve anything.

Each Wisp includes $60 of AI usage a month, with a daily budget that acts as a hard cap. When it is spent, Wisp keeps updating from what it already knows and the model pauses until the next day. You are told once. If you want more, add another month’s allowance for $60, whenever you choose. Nothing is ever added to your bill automatically.

Not yet. Wisp is hosted only because we answer for it: it wakes on events, reaches phones, enforces budgets and holds an authority ceiling, and we cannot stand behind that on servers we cannot see. Self-hosted installs show Wisp locked, not hidden, with the reason, and you can ask below to be told if that ever changes.

Watching and triage: continuous watching across performance, security and tickets, nudges to the right phone, the brief, one durable thread per person with links to the tickets, quotes, invoices, devices and alerts it is about, memory you can confirm or retract, the daily budget and spend meter, pause and a proactivity dial, and bounded work handed to the specialist agents. The setup designer, where Wisp proposes a customer’s monitors, routing and ticket rules for your approval, comes later. We price what ships.

You do. Wisp works under your technicians’ authority and inside your risk engine, and every action it takes is in your audit log. Nobody at Breeze watches your queue for you. The Supervised Rollout is optional: thirty days with the people who built Wisp, setting it up with you and teaching your team to supervise it. It comes with a founding member’s first Wisp.