Our vision
An assistant suggests. An agent does.
The difference shows in what happens next: if someone has to copy the result across by hand, it was not an agent. Ours create deals, send quotes, call prospects back and answer customers — and leave a trace of every move.
No credit card required
The conviction
What an agent has to be before you trust it with anything
It announces itself
An agent that passes for a human wins a conversation and loses a customer. Ours introduce themselves as what they are. It is a matter of honesty, and soon of law: we may as well do it because we believe in it.
- Identified as an agent to the person it speaks to
- Every message attributed to its author
- The transcript stays available
- No impersonation of a human
Its autonomy is a setting
Two modes, and you choose: autonomous, the agent answers and acts; copilot, it drafts and a human approves. The setting is per channel — the same agent can run chat autonomously and email in copilot.
- Autonomous or copilot, your call
- Set per channel, not globally
- Human takeover in one click, at any moment
- Handed back to the agent once the question is settled
It draws on what you know
An agent plugged into nothing makes things up. Ours are grounded in your knowledge base: your documents, your pages, your rules. And when no relevant passage is found, the product says so rather than letting you believe it knew.
- Answers cited from your own base
- A missing source is flagged
- One base for chat, phone and agents
- Corrected once, up to date everywhere
It is paid for when it works
AI is billed on actual use, in credits. No compulsory AI subscription, no “per resolution” pricing that charges a package whatever happens. An agent that does not work costs nothing.
- Credits drawn on actual use
- No AI package imposed
- The cost of each use is shown
- The model is chosen per use, not fixed
In practice
What that gives you in the product
FAQ
Frequently asked questions
Can an agent act without my knowing?
No. Every action goes through the product and leaves a trace there: who did what, when, and on what basis. That is what makes it possible to grant autonomy without losing control.
What happens when the agent does not know?
It hands over with the full context, and the human receives a draft answer to review. It does not improvise something plausible — that is the difference between a grounded agent and a conversation engine.
Do agents speak several languages?
Fourteen, Japanese and Korean included. The agent replies in the visitor's language, and when a human takes over, translation continues both ways: the customer never sees the relay.
Can one agent drive another?
That is the whole point of the architecture: one service can orchestrate the others. The main agent, and how much autonomy you grant it, shape your experience of the product.
Another question? Write to us — a human will answer.
Put an agent to work
Start in copilot on a single channel. You will open up autonomy once trust has been earned.
No credit card required