Do we need to change the core system, or get its vendor's help?

No. Kitewell uses the same screens and accounts as a person. It never touches the core system's programs or settings.

Won't it stop when the core system's screens change?

When a button moves or a field is renamed, AI reads the screen again, relearns the operation, and carries on. The run record shows what it fixed. If AI cannot fix it, a person is told where it stopped.

Could it register something by mistake?

A person can approve before anything is registered or updated. A step can also read the screen again after a write to check it went in correctly. Every run is recorded with screenshots.

How many items can it handle?

It works at a person's pace, without breaks. For more items, add computers that run Kitewell and process in parallel.

On your servers and PCs, in closed networks, and on VMs in your Azure, AWS, or Google Cloud. It reaches core systems that exist only inside your network.

Quoted by the number and scale of the systems to connect. The check of whether it can connect is free.

Told it can't be connected because there's no API? We connect it.

Name: Demonstration: an AI call center connected to a customer system with no API (39 seconds, silent)
Uploaded: 2026-10-22
Duration: 39 s
Description: An address change taken by an AI call center is entered in a customer system with no API, and the receipt number goes back. Midway, the customer system switches to its redesigned screens, and the unchanged workflow keeps running.

Core systems, old web systems, and Windows apps. Kitewell's AI RPA works their screens the way a person does and makes them callable from kintone, Salesforce, and AI. No change to the core system, and no API project from its vendor.

Get a free connection check Download the overview

See it work

An address change taken by an AI call center is entered in a customer system with no API, and the receipt number goes back. Midway, the customer system switches to its redesigned screens, and the unchanged workflow keeps running.

Real Kitewell runs against stand-in systems built for the demo, recorded on October 22, 2026, with Japanese screens. Every request arrived by webhook.

Was a connection like these turned down for lack of an API?

A partner approved in kintone
Registered in a sales system with no API
Invoices read by AI-OCR
Entered on the accounting software's screens
An address change taken by an AI call center
Registered on the CRM's screens
Orders in Salesforce
Entered in the core system, with the order number sent back
Stock in the core system
Sent to the online store and Slack every morning
Questions from Claude or ChatGPT
Answered by looking up stock and orders in the core system
Orders on a partner's Web-EDI
Registered in the order system
Reports from a Windows business app
Exported to Excel or BI

Real examples

Why others turn it down

Integration through the screen breaks whenever a screen changes. Nobody can tell how much fixing it will take, so most companies decline it or recommend rebuilding the system. Kitewell's AI fixes what breaks. That is why we can take it on.

Compared	People re-key it	Have an API built	Hand-built robots	Kitewell
Changes to the core system	None	Required, through the vendor and at a cost	None	None
Getting started	Right away, but people every day	Wait for a quote and development	Assemble scenarios by hand	Say what you need, and AI builds it
When a screen changes	People adjust	Unaffected	Stops until its builder fixes it	AI fixes it and records why
Called from outside, returning results	No	Yes	Needs another mechanism	Called by webhook, API, or MCP, and returns the result
Work left for staff	Daily re-keying	None	Fixes every time it stops	Only approvals and exceptions

AI RPA turns a screen into an API

Kitewell turns the screens of a system with no API into a part other systems can call.

Callable from outside: kintone, Salesforce, AI call centers, and your own systems call it by webhook or REST API. AI agents such as Claude and ChatGPT call it over MCP.
Learned steps, repeated reliably: An operation it has learned replays the same way with no AI. Every run gives the same result, so it can write to core systems.
AI fixes a changed screen: When a button moves or a field is renamed, AI reads the screen again and relearns the operation. What it fixed is recorded.
Results sent back: The number it registered or the value it looked up goes back to the calling system over HTTP, and to Slack or Teams.
Runs next to the core system: It runs in your network, a closed network, or your own cloud, so it reaches core systems that exist only inside.

Requests

kintone, SalesforceApprovals and orders by webhook
AI call centersWhat a caller asked, by API
AI agentsLookups over MCP

Your network

KitewellLearns the screen work, replays it with no AI
Core systemScreens with no API
StaffApprove before a write

Results

The calling systemOrder numbers and lookups over HTTP
Slack, TeamsResults, and what needs a check

Careful, because it touches core systems

No changes to the core system. It uses the same screens and accounts as a person
A person can approve before anything is registered or updated
Every run is recorded with its operations and screenshots of the screen
AI never sees passwords; they are encrypted on the computer that runs the work
It runs in your network or a closed network, with the AI you choose

Where it runs and which AI it uses

What it connects

Core systems such as sales and production management
Old internal web systems
Windows client apps
Terminal emulator screens
Accounting and payroll software
Partner portals and Web-EDI
Government websites

A free check of whether it can connect

Tell us the system to connect and what to connect it to. We usually reply the same business day, and after looking at the screens, we tell you whether and how it can connect.

It costs nothing
You do not need the system's exact name
Send screenshots of the screens after we reply

How it goes

Once it is built, your team can fix it.

1
Free check
We learn the system's screens and what to connect it to, and tell you whether and how it can connect.
2
Connect it on the real screens
In a PoC, the integration runs on the same screens and data as production, to show it works reliably.
3
Into production
Decide where it runs, who approves, and who is told, and Descarty helps you move into production.
4
Your team can fix it
AI fixes a changed screen. To change the steps, staff talk to AI. The reason for each fix stays in your knowledge.

For AI and SaaS vendors

We take on the deals that stall because a customer's core system has no API. Your product calls Kitewell, and Kitewell connects your customer's core system.

OEM and embedding Partners

Frequently asked questions

Do we need to change the core system, or get its vendor's help?: No. Kitewell uses the same screens and accounts as a person. It never touches the core system's programs or settings.
Won't it stop when the core system's screens change?: When a button moves or a field is renamed, AI reads the screen again, relearns the operation, and carries on. The run record shows what it fixed. If AI cannot fix it, a person is told where it stopped.
Could it register something by mistake?: A person can approve before anything is registered or updated. A step can also read the screen again after a write to check it went in correctly. Every run is recorded with screenshots.
How many items can it handle?: It works at a person's pace, without breaks. For more items, add computers that run Kitewell and process in parallel.
Where does it run?: On your servers and PCs, in closed networks, and on VMs in your Azure, AWS, or Google Cloud. It reaches core systems that exist only inside your network.
What does it cost?: Quoted by the number and scale of the systems to connect. The check of whether it can connect is free.

Get a free check of the system to connect

Tell us the connection you were told needs an API, and we tell you whether and how it can connect.