NrtVixAIAGENT CONTROL & SECURITY
Product · Capabilities

Everything your agent can do.

An NrtVix agent does far more than refunds. Here is the full picture: what works in the current build, and what is planned, marked clearly.

NrtVix AI / Product / Capabilities

Watch it work

One conversation. Four kinds of work.

The agent answers, admits what it does not know, takes an action inside your limit, and asks before a sensitive one.

One agentAnswers, actions and approvals in one conversation

Converse

A conversation that feels natural, and stays accurate.

Natural, streaming answers

Replies appear as they are written, in clear professional English, formatted with lists and links.

Remembers the conversation

Recent turns stay in mind, so a follow-up like “and their phone number?” makes sense.

Says “I don’t know”

Business questions are answered only from your information. When the answer is not there, the agent says so.

Resists manipulation

Customer messages are data, not instructions. “Ignore your rules” can at most make the agent ask.

Know your business

Your policies, in your words.

Policies and FAQ

Refund, return and shipping rules, opening hours, common questions.

Products and prices

What you sell, what it costs and what is included.

Per-agent knowledge

Each agent has its own business information, so a sales agent and a support agent can differ.

Files and website pages Planned

Upload documents or point at pages instead of pasting text.

Take action

Real work, through checked tools.

Look up customers

Find a customer record by reference and summarise it.

Refund orders

Inside your automatic limit at once; above it, only with your approval.

Add ticket notes

Record what happened on a support ticket for your team.

Prepare data exports

Customer exports, limited by row count and reviewed when large.

Refunds through a payment provider Next

Real refunds, starting in the provider’s test mode.

Orders on store platforms Planned

Order status, changes and cancellations.

Email and messages Planned

Draft and send replies from your support inbox.

Stay in control

Guardrails that do not depend on the model.

Three clear outcomes

Every action is allowed, sent for your approval, or blocked.

Exact approvals

You approve the specific request you were shown; a changed request needs a new approval.

Budgets

A monthly limit per agent. Over budget, money actions stop.

Emergency stop

One switch halts every agent and cancels outstanding permissions.

Tamper-evident audit

Each entry is chained to the one before, so a silent edit can be detected.

No keys for agents

Credentials stay with the trusted gateway and are never shown to an agent.

Work your way

Hosted, connected, or both.

Hosted agents, no code

Create, teach and chat with an agent from your browser.

Your own agents Preview

Route tool calls from an agent you built elsewhere through the SDK.

Research, Business and Coding agents In development

Specialist agent types beyond customer support.

Website chat widget Planned

Let customers talk to your agent on your own site.

Desktop apps Planned

macOS, Windows and Linux, with notifications when a request needs you.

Abilities without a label work in the current build on sample data, and connect to your real store during early access. Labels such as Planned, Next and In development mean exactly that.

Early access · December 2026

Let agents act. Keep control.

We are opening NrtVix AI to a small group of businesses first. Tell us about yours and we will be in touch.