We're building AI to replace customer service jobs.
You read that right. It's going to happen, whether we build it or someone else does. We can't change that. What we can do is decide where the value goes.
So we made a rule. Either 5% of our qualifying revenue or 70% of our adjusted profit, whichever is larger, goes to the people whose jobs we automate. From our first dollar, not our first profit. And we're building toward giving every one of them a real stake in the company itself.
Starting now.
If we want things to be different, we must do things differently.
Whatever support looks like at your size.
Your team answers the same handful of questions all day long. Let Celeste take those, and keep your people for the ones that need a person.
Volume is climbing faster than you can hire. Celeste takes the overflow, and the team you already have keeps up.
You answer every email yourself, nights and weekends included. For $49 a month, Celeste is the support desk you could never justify hiring.
Don't flip a switch. Turn a dial.
Celeste starts in Practice, drafting on your real tickets in private while we score each draft against what your team sent. When a topic is good enough, you promote her, one topic at a time.
How autonomy worksCeleste learns.
Then you try.
Then you trust.
Trains like a new hire
Celeste reads your team's actual replies, the training documents you give her, and any guidance you type in.
Works a probation
In Practice, Celeste drafts on your real tickets in private, scored against what your team sends. No customer sees a thing.
Earns autonomy, per topic
When a topic is ready, you promote her. The easy ones go automatic while the sensitive ones stay supervised.
Stays transparent and guarded
Every reply shows its sources and confidence. Every risky action is held to the limits you set.
Fair questions. Straight answers.
What if Celeste makes things up?
AI can make mistakes, so we built for that. Celeste backs every claim with your own business knowledge, and an answer she can't support from it is held for a person instead of sent.
What if Celeste refunds the store away?
You set plain limits like "never refund over a thousand." Limits run as deterministic code outside the AI model, so no prompt and no model judgment can override a gate on a tool Celeste can use.
How tool safety worksWhat if I can't tell why Celeste acted?
Nothing is a black box. See what she retrieved, how confident she was, and why she acted, on every reply.
What if answering isn't enough?
Celeste answers knowledge and process questions, and she can also run tools you give her access to, your own APIs or MCP servers, to look up an order, check eligibility, or take the action, inside your limits.
The AI proposes. Hard code decides.
Your support history is only ever used to answer your own customers, never to train a model. And every override and change is logged, so nothing happens behind your back.
Read about securityOne clear price per resolved conversation.
Monthly plans, with a batch of conversations included. You pay when Celeste answers, and nothing when she hands off or stays quiet. Set your own ceiling and she won't cross it.
Every plan starts with a free one-month trial. No card to start, and you're not charged until you choose a plan.
Start your free monthWebsite live chat is metered on its own, with a separate included allowance and a per-resolution rate that starts a little higher than email. Per-plan web chat rates are on the pricing page.