Hiulix
Talk to the agent
Pre-launch Pilot slots open for MSPs — run by the founder himself

The most intelligent first reply
an MSP can put on its website.

Hiulix answers the moment a prospect lands — a dead file server, a provider they have had enough of — asks the five questions you would ask, writes the lead record while they type, and routes it to the person who should call. At 3am. On a Saturday.

Talk to the agent See pricingFrom $199/mo

This is the real product, running on our own site. Bring your own MSP and argue with it.

app.hiulix.com/inbox
Visitor Live Intent: provider switch
Hi — I’m an AI agent, not a person, so you won’t be waiting on anyone. What’s going on?
Looking at IT providers. We’re a dental group — 60 staff, two offices.
Understood. Two questions and I can get the right person on this: is there anyone internal handling IT today, and when does your current agreement run out?
Nobody internal. The agreement’s up end of January.
Introducing Hiulix

Most website chat was built to deflect support tickets. MSPs point it at sales anyway, so it answers three FAQs, asks for an email, and calls that a lead. Meanwhile the buyer on the other side has already changed how they shop — a majority of B2B software buyers now start vendor research inside an AI chat window rather than a search engine. The business that just lost a server at 9pm is not filling in a form.

Hiulix is built for one job and one kind of company: the first sales conversation on an MSP’s website. It works out whether the visitor is a 20-seat prospect or a 250-seat one, asks what your account manager would ask, writes the lead record while they type, and routes it to the person who should call. When it hits the edge of what it knows, it stops and gets a human — which is the part most of this category still gets wrong.

An empty desk by a window in a quiet office, laptop open

The business whose server died at 9:40pm on a Friday is the most qualified lead an MSP will see all month. It is also the one most likely to reach a contact form.

0enquiries missed overnight
<5sto a first, useful reply
5qualifying answers before anyone picks up
An AI sales layer for MSPs

Anything your best account manager does in the first ten minutes, Hiulix does at 3am.

Not a broad platform configured down into a sales bot, and not a horizontal product that happens to tolerate MSPs. One product, one trade: it engages, reads intent, answers from your approved material, qualifies on seats and agreement dates, collects, recommends, schedules, routes and records — and does nothing else.

“Do you cover practices our size?” Exploratory
“What would 40 seats cost?” Pricing
“Our contract ends in 30 days.” Qualified

It reads intent, not keywords.

Every message is classified — informational, exploratory, sales-qualified, pricing, demo, support, or needs-a-human — and the conversation takes the path that matches.

Three colleagues working together at a table with laptops

It asks the five questions you would.

Seats or endpoints. Internal IT, or none. Current provider, and when the agreement ends. The compliance driver — SOC 2, HIPAA, CMMC, cyber insurance. And why now. One at a time, inside real answers — never a form dropped on a stranger.

Assessment booked · Thu 10:30 auto
Lead record written auto
Dan notified · 60 seats auto

It finishes the job, then gets out of the way.

Books the assessment, writes the lead record, notifies the right person, logs the whole exchange for analytics. Whoever calls opens a lead that already has the seat count, the agreement date and the trigger attached.

Why now

A category compounding faster than anyone can staff against.

Analysts disagree on the size of the conversational AI market by several billion dollars, because they define it differently. They do not disagree on direction. Here are five 2026 estimates, side by side, rather than the single flattering one.

GV Grand View Research
—
PR Precedence Research
—
IM IMARC Group
—
RM Research and Markets
—
CM Coherent Market Insights
—

Base years are 2025 or 2026 and forecast horizons run 2032–2035, depending on the firm — the bars compare magnitude, not identical periods. North America is consistently put at 28–31% of the global market. The gap that matters more to us is smaller and better measured: the U.S. Census Bureau puts small-business AI adoption near 20% under a strict definition, while broad self-report surveys reach the 70s. Most of that reported adoption is someone pasting text into a general chatbot — not a system doing a job.

Knowledge layer

Knowledge that only says what you approved.

The agent answers from sources you load and can see — your service catalogue, your coverage area, your minimum seat count, your qualifying criteria. Scoped to your account and never mixed with another MSP's.

  • It says “I don't know” out loud. Off-script questions become a handoff, not a guess.
  • Every source is switchable. Turn a page off and the agent stops quoting it on the next message.
  • One account, one knowledge base. Nothing you upload is shared across customers.
  • Model-independent by design. The orchestration layer sits between your business and whichever model runs underneath, so the provider can change without the product changing.
Sales intent engine

Intent that routes itself.

The same visitor question means different things at different moments. The engine decides which of seven paths a message belongs on, and the workflow engine acts on it. It sounds like this: “What do you charge per seat?” → pricing

Informationalanswer
Exploratoryqualify
Sales-qualifiedroute
Needs a humanescalate

Seven paths, one conversation.

Informational, exploratory, sales-qualified, pricing, demo, support, needs-a-human. The visitor never sees the machinery — they just get an answer that fits what they asked.

A sales team at a shared table, several on headsets

Your team gets context, not a ping.

A routed lead arrives with the seat count, the current provider, the agreement date and the trigger already captured — so the first human call starts where the chat ended.

Conversations312
Qualified leads47
Escalation rate9%
Median response4s

Numbers you can actually argue with.

Conversations, qualified leads, conversion events, escalation rate, response time, usage. Pilots that fail usually fail because nobody could measure them. These are the measures.

Human handoff

The best thing it does is stop.

A visitor who asks for a person, gets frustrated, or asks something outside the knowledge base gets a human — immediately, or a logged follow-up if it is the middle of the night. It is a default, not a setting you have to find.

Try to break it
Data & control

Private enough to point at client work.

The agent handles your service catalogue and personal details volunteered by strangers, on the site of a company whose own business is other people's security. That makes data handling part of the product, not a page in the contract.

Only what qualifies a lead.

Name, contact details and the substance of the enquiry. Not a broad, undefined sweep of whatever a visitor happens to type.

Encrypted in transit and at rest.

Access is limited to your own authorised users and the few Tora Labs staff needed to run and support the platform.

Retention with an end date.

Conversations and leads are kept for as long as your analytics need them, on a defined period written into the terms — and deleted on request.

Sub-processors on the record.

The AI providers and hosting we rely on are listed and reviewed, so swapping a model provider never quietly changes where your data goes.

Straight answers

The questions people actually open with.

Short versions here. The agent will take any of them further, and will tell you when it does not know.

Do you have customers yet?

No. Hiulix is in pre-launch testing and Tora Labs has no paying customers and no revenue. The founder is validating conversational accuracy, the website embed and the qualification logic before taking money for it. What is open right now is a pilot, run by the founder personally — which is a better deal than being customer four hundred.

Why only MSPs?

Because the first sales conversation an MSP has is the same conversation every time — seats, internal IT, current provider and agreement date, compliance driver, and why now — and an agent can be genuinely good at a conversation that standardised. A product aimed at every industry has to be configured into usefulness by whoever buys it. Tora Labs' five-year plan names IT service providers among its first markets; this is that line taken seriously rather than hedged.

How is this different from the chat widget we already have?

Most website chat descends from customer support and was built to deflect tickets. Hiulix is built around one job — qualifying and routing an MSP's sales conversation — so it asks the five questions on its own initiative, writes a structured lead record, and measures itself on qualified leads rather than deflection rate.

What stops it from inventing things about our services?

It answers from a knowledge base you load and control, scoped to your account alone. When a question falls outside it, the designed behaviour is to say so and escalate rather than improvise. That is the single most important design constraint in the product, because a confident wrong answer to a prospect costs more than no answer.

How long does deployment take?

Days. The customer installs a lightweight website component and loads the knowledge sources; the qualification questions and routing rules are configuration, not a build. Enterprise conversational AI platforms typically run weeks-to-months implementation projects — that gap is the whole positioning.

Is voice available?

Not yet. Voice AI is in development and the platform architecture is being built for it, but it is not shipped and the site will not claim it is. Website chat comes first because it is the controlled environment in which to prove the product works.

Will it integrate with our PSA?

PSA, RMM and CRM integrations are being built out through the first 24 months. Nothing is live today. Ask the agent about ConnectWise, Autotask or HaloPSA by name and it will tell you that honestly rather than nodding along — which is the behaviour worth testing before you trust it with a prospect.

Can we resell it to our own clients?

Not yet, and there is no white-label or per-client-site program to describe. It is the first thing most MSPs ask, and the reason is obvious from both sides — one MSP reaches dozens of SMBs who each have a website. It is not built and not priced, so the honest answer is that it is a conversation with the founder rather than a product page. The published ranges are per MSP.

Who is behind this?

Tora Labs LLC, a Pennsylvania company (entity 0014963635, filed October 30, 2025), founded and wholly owned by Nodirbek Anvarov, who capitalised it with $50,000 of his own money and built the MVP himself. There are no outside investors. More on the company.

Built for MSPs and nobody else. Start with a pilot.

Bring a real question from your own site to the agent and see how it handles it. If it goes well, the next conversation is with Nodirbek.

Talk to the agent

No card, no form, no sales sequence. It is a conversation.