Přejít na obsah
MODULE · MAGIC AGENT

From request to draft quote, while you do something else

It searches, calculates, makes the proof and prepares the quote. Your team looks at it and sends it.

Co to je

O čem to je

Magic Agent is not a chatbot that answers but an agent that performs actions. A request arrives through WhatsApp, the widget on your site, email or a form. The agent understands what is asked in plain language, finds the article in your own catalogue of 74,136 products, picks the variant, applies the decoration tier, has Magic Logo make the proof and prepares the quote in the system.

Your team looks at it and sends it. The agent can also check stock, arrange a sample order and recognise a repeat order — a repeat order has no new set-up costs. At the Brinxx group it has been preparing around 380 quotes a month since 2025, across three brands.

Uw klanten communiceren via WhatsApp of de web widget. Ze vragen een offerte aan in gewone taal "ik wil 200 mokken met lasergravure" en de agent regelt de rest. Logo uploaden, achtergrond verwijderen, artwork op het product plaatsen, offerte genereren met staffelprijzen en setup kosten, en het geheel als PDF mailen. Alles binnen 2 minuten. De klant schrijft in het Spaans. U werkt in het Nederlands. De agent vertaalt alles on-the-fly. Geen misverstanden, geen vertraging. Maar de agent doet meer dan offertes. Hij verwerkt sample orders, checkt voorraad bij uw leverancier, herkent producten op foto, zoekt in natuurlijke taal ("iets roze waar je uit kunt drinken"), en signaleert return orders zelfde configuratie eerder besteld? Geen setup kosten. Op de backoffice genereert de agent offertes met AI, laat u regels per stuk goedkeuren, en pusht goedgekeurde orders via de Magic Connect naar uw productiesysteem of dat nu FastEditor, uw ERP of een ander systeem is. Van web naar machine, zonder tussenstappen. En als de agent het niet weet? Dan escaleert hij naar uw sales team via WhatsApp in de taal van uw medewerker. Geen klant gaat verloren. Hoe houden we de AI betrouwbaar? Met de Flowbuilder. Elk gesprek volgt een voorgedefinieerd proces. De AI interpreteert alleen wat de klant bedoelt de flow bepaalt wat er gebeurt. Geen hallucinaties. Geen onverwachte antwoorden. Elke stap is voorspelbaar en controleerbaar.

Magic Agent werkt op twee niveaus: op de frontend voor klantinteractie en op de backend voor automatisering en systeembeheer.

Pro koho

Pro koho jsme to postavili

Companies where requests arrive through four channels and the first response takes too long. And companies working in several languages: the agent works in plain Dutch, English and German and translates along the way.

Co s tím máte doma:

Offertes maken met artwork preview in 2 minuten
Escalatie naar sales team in meerdere talen
Sample orders volledig geautomatiseerd
Product zoeken op foto met camera
Return orders herkennen — geen setup kosten
Flowbuilder stuurt elk proces — AI beteugeld
Offertes goedkeuren per regel
Orders pushen naar FastEditor — productie-klaar
On-the-fly vertalingen in 5 talen
Voorraad checks via PIM integratie
WhatsApp en web widget integratie
AI product zoeken in natuurlijke taal
Co to dělá

Co se v praxi děje

The work the agent takes over is not the conversation but the preparation: searching, calculating, making the proof and setting up the lines. For nine out of ten companies in this industry that is the same action, dozens of times a week.

Preparing quotes from any channel

From email, the widget or the form on your site, with the artwork included rather than attached afterwards.

Finding products in plain language

Including variants and alternatives, and also from a photo: a customer sends an image and the agent finds the closest article in your catalogue.

Calculating decoration

With the right technique, tier, number of colours and set-up costs from the 426,573 technique rules in the PIM.

Making a proof through Magic Logo

And sending it along, so the customer sees what they are getting instead of having to imagine it.

Checking stock and lead time

In the PIM, before promising anything. A lead time the agent cannot support, it does not promise.

Recognising repeat orders and samples

A repeat order gets no new set-up costs, and a sample request follows its own route.

Approval per line

Your team approves or rejects line by line, and the order is made from the approved lines. A customer choosing between two variants does not need a new quote.

Escalating to a human

If it stalls or gets delicate, an employee takes over the conversation, even mid-way through a WhatsApp chat.

Součásti

Centraal AI controlepunt

Magic Agent is het brein dat alle onderdelen van MagicEverse met elkaar verbindt. Van klantgesprek tot backend automatisering.

01

Data Access

Volledige toegang tot PIM, orders, klanten, producten

02

Escalatie Server

Ingrijpen bij escalaties, WhatsApp sessie overname, 2FA verificatie

03

Offerte Generator

Maakt offertes direct vanuit gesprek of trigger

04

Customer Manager

Creëert en beheert klantaccounts

Souvislost

Jak to navazuje na zbytek

The agent reaches the PIM, the quote module, the ERP and the mailbox through the MCP servers in Magic Orb. The conversation runs through Widget Server and, for WhatsApp, through Meta Server. The logic sits in Flowbuilder. The proof comes from Magic Logo, the order goes to the ERP and on to Fast Editor if needed.

The line we hold is the same on every channel: the agent prepares, the human decides. Nothing reaches your customer without approval, you can always see why the agent did something, and we do not use your business data to train third-party models.

Co to přináší:

Begrijpt natuurlijke taal in NL/EN/DE
Interpreteert intentie, niet alleen keywords
Voert complexe taken autonoom uit
Schaalt van 1 klant tot bulk operaties
± 380
quotes a month at Brinxx, Promotionalz and Moxz
74,136
products the agent searches
426,573
technique rules it calculates with

Send us last week's requests and we will let the agent loose

You get back what the agent made of them: the articles found, the decoration calculated and the draft quote. That shows you not whether the technology works but whether it works on your own requests.

You will also hear what did not work. A request the agent cannot handle is more useful to see than ten it can.