Build your business software with AI, on blocks it cannot route around.
Core is an AI coding platform. You build your order processing, stock control, invoicing and customer portal with Claude, ChatGPT, an open model or by hand. Whoever builds, it connects blocks that each do one job, and you decide what goes live.
execution · order #29,107
- 00:00.000webshop.order · order #29,107 received
- +180 nsvoorraad.check · 3 lines reserved
- +1.2 µswms.picklijst · station 4 instructed
- +2.1 µsexact.factuur · posted in Exact Online
- +0.4 msmail.tracktrace · customer notified
- ✓ doneflow executed · 6 blocks in 0.4 ms
avg. per block214 ns
Why AI cannot take a shortcut on Core.
Core works as a container for what AI builds. The model picks blocks and connects them. What a block may do, how it is secured and what goes live is settled by the platform, and by you.
- Only blocks from the library
- The model builds with 2,800 blocks we write, test and maintain, each with one job and fixed inputs and outputs. A flow has no place for loose code that works around them.
- Permissions live on the block
- Who may call an endpoint, and how often, is a setting on the block itself, visible to anyone looking. There is no line 412 where a check quietly went missing.
- Nothing leaves unseen
- When a flow calls another system, there is a block for it on the canvas. Requests to internal addresses are stopped by the platform unless you allow them, and they land in the audit log.
- You publish
- What the model builds is a draft. Studio checks the connections and required fields, nothing goes live until you publish, and every change can be rolled back.
In the assistant's wordsBuild an endpoint that takes in webshop orders.
- Changes: Webshop orders
- Tests stay inside the cluster.
- Nothing goes live until you publish.
- Blueprint: Webshop orders
- Draft: 16 new blocks.
- This goes live as soon as you publish.
⌘ Enter to answer
Three ways to build, with the same blocks.
Which one you pick makes no difference to the result: the same blocks, with the same checks.
The assistant in Studio
Runs on Claude Code or Codex, on your own Claude or ChatGPT plan and the model you choose. You say what a flow should do and it puts a draft on your canvas.
See the assistant
Your own agent
Claude Code, Claude Desktop or another agent that speaks MCP, including one running an open model on your own hardware. It works through your session, past the same checks, and every action lands in a log.
See agent access
By hand
In the Flow Editor you drag and connect blocks yourself, or edit a flow as Blueprint Language in the code view. It is also how you check what AI built.
See the editor
In plain language: how it works
Three layers, in this order. Once you have seen it, you see it everywhere.
Your data lives in tables
Orders, customers, stock, machines: everything gets its own table, laid out the way you already picture it. No database knowledge required.
Flows do the work
A flow is a series of blocks: an order comes in, check the stock, create the invoice, print the label. You decide the steps, Core runs them, day and night.
Screens for your people
On that same data you build the screens: a picking screen for the warehouse, a dashboard for the office, a portal for your customers. Web and mobile.
And when your process changes next month? You ask the assistant, or drag a block into place yourself, and publish. No release cycle, no queue at a software vendor.
See the integrationsFaster than your database can blink.
Core is not a layer on top of someone else’s cloud. We wrote the engine ourselves, in a compiled systems language, and it runs blocks in nanoseconds. That is why a flow with dozens of steps finishes in under a millisecond, and why a single environment handles millions of automations a month. What AI builds runs just as fast: it is the same blocks.
You will not notice that in a benchmark. You notice it in a picking screen that never waits and a sorting line that never stalls. Below is a measurement on a production cluster. For comparison: a blink of an eye takes 100,000,000 nanoseconds.
execution times · production cluster
reading a field from a record
31 ns
running a calculation
64 ns
saving a record, with audit log
2.4 µs
a full order flow, 40 blocks
0.4 ms
For comparison: a blink of an eye lasts 100,000,000 ns.
What is in that monthly price?
A Core licence does not replace one program, it replaces the entire stack underneath it. This is what you never arrange yourself again.
Hosting, DevOps and monitoring
Dutch data centres, automatic migration when something fails, 99.9% uptime over the last 5 years. None of it lands on your desk.
Security that is simply on
Threat protection, encryption, audit logs and hourly backups. Built into every environment from day one.
Daily updates, zero maintenance
We maintain the platform codebase. Your application is assembled from blocks, so it inherits every improvement without a rebuild and without technical debt.
Connections to your whole stack
Exact, Mollie, Floriday, PostNL, Anthropic and dozens more, as ready-made blocks. Plus webhooks and APIs for everything else.
Put that next to an in-house development team, servers, security tooling and an integration partner. Licences from € 515 a month.
See all licencesOnly need one app, built by you?
Core Nano is the same engine in one small box, for the solo builder: database, accounts, deployment and the Flow Editor, for €25 a month, fixed.
Further into Core
Where you build, what you build with, and where it runs.
Studio
The workshop around the platform: an assistant on Claude or ChatGPT, together live in the same flow, with the block library, comments and deploy management.
See Studio
The Flow Editor
The canvas itself. Connect blocks, watch real data run through them and roll back any version.
See the editor
Hosting
Dutch data centres, racks across several floors and networks, automatic migration when something goes down.
See the hosting
Watch AI build it live, with your own process in it.
In a one-hour demo we build a first flow for your operation, with AI and by hand. Actually build it, not show slides.