Services Work Research Product About Contact
Our Product · AI Employee

Moj Kolega (My Colleague)

An AI employee for your webshop. It answers customers, recommends from your catalogue, captures the leads you'd lose after hours, and hands off to your team when it should - 24/7. We set it up, watch it, and improve it every month. You see results, not another chatbot.

Visit Moj Kolega Build Something Similar
mojkolega.hr - the Moj Kolega homepage, with the agent answering a real catalogue question in the embedded chat widget An overnight report - questions answered, recommended and escalated through the night, resolved alone vs handed to the team The five jobs Moj Kolega does - grounded answers, recommendations, lead capture and nine languages, all from the shop's own data
24/7
Answering customers, day and night
9
Languages it detects and replies in
5
Jobs it does, grounded in your data
Monthly
We tune it, every improvement cycle

Hire a Role, Not a Chatbot

Most shops bolt on a chatbot, watch it guess wrong, and switch it off. Moj Kolega is the opposite: a managed AI employee that answers only from your catalogue and policies, says "I don't know" instead of inventing, and gets better every month because someone is actually watching it.

Grounded, Never Invented

  • Answers only from your catalogue and policies
  • When the data isn't there, it says so and escalates - it doesn't guess
  • Product picks come with a reason to buy, from your catalogue only
  • Every answer is traceable to the source it came from

Managed, Not DIY

You don't get a prompt editor and a model picker. You hire a role - Sales, Support or Reception - and we do the rest: set it up, connect your catalogue, and test it against real questions before it goes live.

Then every month we send a report of what your customers asked, what the agent handled alone, what it missed, and what we changed. The improvement is the product, not a dashboard you have to babysit.

Five Jobs, Done Reliably

Not forty jobs done badly by a generic model. Five that matter for a webshop, each grounded in your own data.

01

Answers, Grounded

From your catalogue and policies. When it doesn't know, it says so and hands off - it never guesses and costs you a sale.

02

Recommends the Right Product

Two or three picks for the need and the budget, each with a reason to buy - drawn from your catalogue, never invented.

03

Captures Leads You'd Lose

Wholesale, B2B and undecided buyers - qualified, with the full conversation, on your desk by morning instead of gone by dawn.

04

Escalates Safely

When it's out of its depth, it hands off to your team with the full context - a clean handoff, not a bluff.

05

Speaks 9 Languages

Detects the visitor's language and switches on its own - Croatian, English, Slovenian, Serbian, German, Italian, Bosnian, Hungarian and more.

06

Works the Night Shift

The hours your team is offline are when leads slip away. Moj Kolega covers them - and hands you the night's conversations in the morning.

One Script, Grounded in Your Shop

Moj Kolega drops onto any site as a single widget - but the work behind it is what makes it trustworthy.

An Embeddable Widget

One script tag on any site. It streams answers token by token, knows which page the visitor is on, matches their language on every message, and can be switched on or off remotely - no redeploy.

  • Streaming chat over SSE, page-context aware
  • Per-message language matching and configurable appearance
  • Remote on/off and roll-out without touching the host site

Knowledge From Your Catalogue

Crawls your site and walks your WooCommerce or Shopify catalogue, plus uploads and text. Hybrid vector-and-keyword search finds the right passage, with product and price extraction that respects EU Omnibus pricing.

Duties: Work in the Background

Beyond the chat, the agent runs scheduled duties for the shop - catalogue audits, monitors, a weekly commerce analysis, even write-backs to your store - each one drafted for review before anything actually changes.

Hard to Trick

Untrusted web and catalogue content is sanitized and fenced, an input tripwire catches injection attempts, and an output scanner stops leaks mid-stream. It won't be talked into going off-script.

Built On

Next.js 16 and React 19 on the front end; FastAPI with async SQLAlchemy on the backend and agent runtime; Celery and Redis for crawls and scheduled duties; PostgreSQL 17 with pgvector for storage and hybrid search - all in Docker, bilingual in Croatian and English. The interesting engineering isn't the model; it's everything that keeps the model honest.

Meet Your Webshop's AI Employee

Moj Kolega is live and taking on shops now. See it answer real questions in your language, then get one set up for your own catalogue.

Visit mojkolega.hr Talk to Us