Manager
Detects language and intent, answers simple messages directly, and routes everything else to one specialist — or straight to a human.
Platform
Detoximex is not a single prompt behind a chat bubble. It is a bounded system where the model proposes and the platform decides.
A manager agent reads the conversation and routes each message to exactly one specialist. A customer message never costs more than two model calls, so response time and cost stay predictable as volume grows.
Detects language and intent, answers simple messages directly, and routes everything else to one specialist — or straight to a human.
Collects the fields your business defines, one or two questions at a time, and reports objective signals about buying intent.
Retrieves from your own documents and answers only from what was retrieved. No match means an honest handover.
Presents real availability, interprets the customer's choice, and lets the platform confirm against the calendar.
The model never decides the score. It reports signals — budget stated, timeline within 30 days, appointment requested, financing ready — and the platform applies your fixed point rules. The same conversation always produces the same score, and you can explain every point to your team.
Qualification fields and point rules are data, not code. Change what matters and it takes effect on the next message.
Rules such as “score at or above 70 alerts a senior agent” live in a policy table you control, not scattered across automations.
Every agent run records the model, version, tokens, cost, latency and outcome — so quality and spend are measurable from day one.
Availability is read from the assigned agent's calendar and re-checked at the moment of confirmation. If the slot was taken between the offer and the customer's reply, Detoximex apologises and offers the closest alternatives instead of creating a conflict.
An explicit request for a human, anger, a complaint, a legal or financial question, price negotiation, or repeated low confidence all stop the AI immediately. Your agent is notified with the lead summary and reason, replies from the workspace, and switches the AI back on when the conversation is done.