Reactive Agent

How Voice thinks on this turn — knowledge, skills, and confirmation. Same product as Voice AI Agent, not a second SKU.

Website and technical copy that says reactive agent means this Voice AI Agent. It is the thinking layer behind Talk: what to say, and which action to run. It is not a separate product you buy on its own.

Reactive here means it acts on this turn. It hears the job, collects what it needs, and can run an enabled skill immediately. It is not a planning agent and is not built for outbound campaigns.

What callers experience

On this turn	What Voice does
Small talk or capability questions	Stays in persona. It does not invent business facts.
A question your knowledge covers	Answers from documents you approved.
A question your knowledge does not cover	Says it does not know — rather than guessing.
A job you enabled as a skill	Collects details, confirms if you require it, then runs that one skill.

Live prices, stock, and similar changing facts belong in a skill, not in a static knowledge article.

Knowledge and skills are yours

Knowledge — approved content per language. Voice is built to stay inside that set.
Skills — actions you write and enable. There is no fixed industry catalog. MinskyAI, an integrator, or your team can deliver a skill; enable is the last step.
Confirmation — optional per skill. When it is on, Voice waits for a clear yes before it runs the action.

Approved tools — including MCP (Model Context Protocol) — can be connected, then enabled the same way. This page does not list tool servers.

Languages and persona

First-class languages are English, Hong Kong Cantonese, and Mandarin. Welcome wording and tone are set per language as part of the agent you run. Talk still owns the audio: listening, speaking, and interrupt.

What this page is not

Apps and websites talk to Talk, not to the thinking layer. Internal agent APIs, routing detail, and operator authoring stay behind a pilot or architecture review.

System view: How it works. Embed: Talk API. Actions: Skill SDK.