AI agents that work inside your systems.
They take over recurring processes in ERP, PIM, and CRM and find the errors and gaps in your data along the way. Your team approves, at a fixed price.
Free: AI search, legal basics, accessibility, and four more checks in 20 seconds, with Claude ranking what to fix first. No signup.
Built for Deutsche Stadtmarketing, ABT Sportsline with Digitize!, and TGOA. 
Live demo
Watch an AI agent at work.
A supplier invoice arrives by email. The agent does the rest: through the interfaces of the ERP and PIM, and where there is none, it clicks through the program like a person.
Sample with invented data. This is how the agents we build into ERP, PIM, and CRM work.
This workflow for your companyHow the demo runs: a supplier invoice arrives by email. The AI agent reads the amount, due date, and items from the PDF. In the ERP it books the invoice through the interface, without a single click. In the PIM it updates the purchase price, also through the interface. The tax adviser's portal has no interface: the agent opens it, uploads the receipt, and fills in the fields like a person. Before submitting, it waits for the team's approval.
Sound familiar?
- Orders arrive as PDFs, and someone types them into the ERP. Business process automation
- Product data lives in three spreadsheets, and nobody trusts any of them. PIM implementation
- The AI pilot worked in the demo. Six months later, it is still a demo. AI agents for business
- The shop, the CRM, and the ERP each know a different truth. CRM and ERP integration
That is the work we take off your team: built into the systems you already run, measured, and handed over running.
What an agent does all day.
Unsorted in, done out. Emails, PDFs, and requests arrive in a jumble, and the agent turns them into checked entries in your systems.
What decision makers ask first?
The objections we hear in every first conversation, and how our agents answer them.
- What if the agent gets it wrong?
- Then someone notices before it counts. The agent records the source of every step, and sending, booking, or deleting goes through approval by your team. Whatever it cannot classify with confidence goes to a person.
- Does it make up answers?
- No. It answers only from your data, with the item number or document as evidence. Without evidence it asks instead of guessing.
- Where does our data go?
- Where you decide: through the model providers' EU data centers or on your own infrastructure. The agent reads only what the workflow needs.
- What does it cost to run?
- The build is a fixed price, from €3.5k. Running costs are model calls per task. We calculate them up front and cap them with a monthly limit you set.
- How do we start small?
- With one workflow, one number before and after, and a fixed price. If the number does not move, we stop.
- What will the works council say?
- That can be checked up front. The agent works on tasks, not on people, and reports show workflows, never the performance of individual employees.
Built into the systems you already run.
Four kinds of work, usually two or three at once. We scope it, build it, and hand it over running.
-
01 · Where we go deepestAI Integration
Agents that read mail and documents, answer requests, and work inside your CRM and ERP.
- Retrieval (RAG) systems built on your own data
- LLM-based agents and internal copilots
- Conversational interfaces and AI-driven automation
- Architecture, build, and production deployment
-
02MDM / PIM Systems
One clean source for your product data, feeding shop, marketplaces, and print.
- PIM platform selection and rollout
- Legacy-database to modern-PIM migration with normalization
- Master-data modeling for products, vehicles, and complex attribute graphs
- InDesign and media-production exports from PIM data
-
03Business Automation
The copying between your tools and the weekly reports, done by software.
- Workflow automation across the tools you already use
- Internal process automation for ops, marketing, and customer ops
- API integrations between business systems
- Monitoring that surfaces breakage before users notice
-
04CRM / ERP Integration
Shop, CRM, and ERP that agree on every order, product, and customer.
- HubSpot integrations: CRM data flows, webshop to HubSpot
- WooCommerce to Billbee product and order sync
- ERP and order-management connectors with reliable failure modes
- Authoritative-system design: where data lives, how it syncs
See it in practice.
- 13 years
- of building software
- RWTH
- M.Sc. computer science, machine learning
- Deutsche Stadtmarketing GmbH
AI, automation, and digital systems for Deutsche Stadtmarketing
AI Integration · Customer Communication Systems · Business Automation
See the work - The Group of Analysts (TGOA)
WooCommerce and Billbee in sync, plus CRM integrations for TGOA
CRM/ERP Integrations · E-commerce · Custom Connectors
See the work - Digitize! GmbH
PIM migration for ABT Sportsline, with Digitize!
MDM/PIM · Data Migration · InDesign Automation
See the work
Who you work with
Hi, I am Philipp.
I have built software since 2013: iOS apps in Krefeld first, then years of product data, including the PIM migration for ABT Sportsline, then shop and CRM integrations. I studied computer science at RWTH Aachen with a focus on machine learning and worked on deep learning for computer vision. Today I build voice bots and automation that run in production at Deutsche Stadtmarketing.
kant.dev is where I do this work for other companies, together with Tobias on analysis and integration. No account managers, no hand-offs: the people in the first call are the people who build your system.
Philipp Kant
Software & AI Architect, Krefeld
Recent posts.
-
What does an AI agent cost? Prices and examples
What does an AI agent cost? Build and running costs at a glance, with model prices (November 2026), a worked example, and the traps where costs run away.
-
Office AI agents: Cowork, ChatGPT Work, and more
Claude with Cowork and Duet, ChatGPT Work, Perplexity Teammates: what the new office AI agents do, what they cost, and where they still fail.
-
Prevent AI hallucinations: answers need evidence
How to prevent AI hallucinations: why agents should answer only with evidence from your data, how that is built, and what happens with contradictions.
