Old in-house systems, unchanged: connected, automated, and with AI.
Core systems, department servers, Windows apps with no import, databases, and file servers. Between systems that live only inside your network, people still re-key data and pass files by hand. Kitewell connects and automates those hand-offs. AI reads documents in any layout, and a person checks before anything is written to a core system. Kitewell runs on your computers and can be set up so business data never leaves.
The gaps between systems that people fill by hand
Are hand-offs between your in-house systems still done by people?
The core system's nightly CSV export
Into the department's Excel and the accounting software
Orders that arrive as PDF, fax, or Excel
Entered in a sales system with no API
Orders in the sales system
Into a shipping Windows app with no import
Stock in an in-house database
A morning stock sheet, by internal mail
Data on the head office server
To the branch servers
Results on the production system's screens
Into reports on the shared folder
Transfers in the HR system
Into the attendance system and payroll software
Orders in an in-house system
Answered on a customer's Web-EDI
See it work
AI reads customers' orders in their own formats (PDF and Excel), a person checks them, and Kitewell enters them in a sales system with no API, then writes the order confirmation and the order ledger. When the sales system is redesigned, AI reads the new screen and keeps entering.
Connect through whichever door each system has
Each system is connected through its most reliable door, and one piece of work can mix doors. Nothing in the core system's programs is changed.
| Door | When | With Kitewell |
|---|---|---|
| API | The system has an API or an integration mechanism | Calls the in-house API over HTTP |
| Files | It can export or import CSV or Excel | Reads and writes files on shared folders or SFTP, and aligns their formats |
| Database | Reading the data is enough | Reads the values it needs with a read-only account |
| Screen | There is no other door | AI works the screen the way a person does, and repeats learned steps without AI |
Kitewell never writes directly to a core system's database. Writes go through import features or the screen, so the business rules apply.
What it connects
Connected and running, even when no one is there
Night and month-end processing runs in the right order, with no one present.
- Starts on time
- At 2 a.m. every night, at month-end close, or when an email arrives: whatever the work needs.
- Runs with no one present
- On an in-house server or a dedicated PC, with no one signed in.
- Keeps the order
- Each system's step finishes before the next begins, so hand-offs stay in order.
- Resumes where it stopped
- A run resumes from the failed step and does not redo finished work.
- Approval before writing
- A person can approve before anything is registered in a core system.
- Records everything
- What went into which system is kept with screenshots, and the person in charge hears when it is done.
What AI does, and what it does not
AI takes the parts people used to judge, and relearns screens that change. Writes to core systems follow checked steps.
- Reads documents
- AI reads orders and invoices in any layout (PDF, fax, Excel), matches them to your masters, and checks the math.
- Builds from words
- Describe the work in plain words, and AI builds the workflow. A person checks it before it is saved.
- Keeps up with screen changes
- When buttons move or fields are renamed, AI reads the screen again, relearns the step, and records what it changed.
- Writes follow checked steps
- Learned steps repeat without AI, and a person approves before registering.
- You choose the AI model
- With your own Azure OpenAI contract or a model on your own servers, screens and documents stay inside. Passwords are never shown to AI.
- AI agents can use core systems
- In-house AI agents, Claude, or ChatGPT can look up stock and orders in core systems over MCP.
Where it runs, and how data is handled
Kitewell runs on your servers, PCs, or VMs. Workflows, run records, and credentials stay on that computer.
Where it runs
- In-house servers or PCs
- Closed networks
- VMs in your Azure, AWS, or Google Cloud
- A VM Descarty hosts for you alone
How data is handled
- Business data, run records, and passwords stay on the computer that runs Kitewell
- Steps that use AI send their content only to the AI model you choose. With an in-house model, nothing leaves the company
Starts
- ScheduleNightly, month-end close
- Internal mailWhen orders or requests arrive
- In-house systems, AI agentsCalled over MCP
Your network
- KitewellOn an in-house server, with no one present
- Core systems and business appsThrough API, files, database, or screen
- The person in chargeApproves before registering
Results
- Shared folders, ExcelReports and ledgers
- Internal mail, chatResults, and what needs checking
Using it alongside the tools you have
Sometimes you do not need Kitewell. Keep using your tools for what they do well.
| Tool | Best for | When connecting in-house systems to each other |
|---|---|---|
| Power Automate and Power Automate Desktop | Connecting Microsoft 365 and cloud services; work a person starts on their own PC | Needs an internet connection, and work-account flows are stored in the cloud (Dataverse). Cloud flows reach in-house systems only through the on-premises data gateway (via Azure Relay), with a paid license |
| Desktop RPA | High-volume, fixed operations on screens that rarely change | Stops when a screen changes, and the builder fixes it. Night runs and work across networks need extra setup |
| EAI and data integration tools (ASTERIA Warp, DataSpider, and others) | High-volume data transfer between systems with APIs or databases | Cannot reach systems that only have a screen, and needs specialists and tool licenses |
| Rebuilding the system | When the business process itself is being redesigned | Takes years and a large budget, and the re-keying continues meanwhile |
| Kitewell | Work across in-house systems that mixes APIs, files, databases, and screens | Runs on your computers, connects through the right door, and AI fixes screen changes |
Japan's Ministry of Internal Affairs and Communications' RPA guidebook for local governments (July 2025) also says RPA is basically deployed per network, and that desktop RPA makes overnight processing difficult.
How an integration runs
The check and seeing it connected are free.
- 1
Check
FreeWe learn the systems' screens and how they work, and tell you which door to use and how to connect.
- 2
We connect it and show you
FreeOn real screens (a test environment, or a recording of the work and sample data), we show data being taken out and put in.
- 3
One-month joint verification
¥500,000 before tax, free if the success criteria are not metWe set the requirements (which data, what starts it, how many, what happens on errors) and the design (which door, where it runs, network conditions, and how credentials are handled), and test it on production screens through errors and screen changes.
- 4
Build and move to production
Building is quoted. Licenses are ¥50,000 per on-premises computer per month before tax. The verification fee is credited to the first yearWe build the rest of the integration, run acceptance tests, and move it to production.
- 5
Maintenance
Maintenance support is monthly (quoted)AI fixes changed screens. When it stops, you are notified and Descarty finds the cause. We agree an incident response time (SLA) and report each month. Your team can also change steps by talking with AI.
Adopting it so your staff build and grow their own automations runs differently.See how AI-native RPA adoption runs
A check of whether it can connect
Tell us which in-house systems you want connected. We usually reply the same business day, and after looking at the screens and how they work, we tell you which door to use and how to proceed.
- You do not need the system's official name
- Networks with no internet access are fine
- You send screenshots after we reply
Frequently asked questions
- Do we need to change the core system, or involve its vendor?
- No. Kitewell uses an API or file import where one exists, and otherwise works the same screens with the same accounts as a person. Nothing in the core system's programs or settings is changed.
- Does it work in a network with no internet access?
- Yes. Kitewell runs on an in-house computer, and its workflows and run records stay there. Steps that use AI can use a model running on your own servers. We go over your environment's conditions in the check.
- Can it connect systems on separate network segments?
- It never makes traffic cross a network boundary. Kitewell goes where it can reach both systems, or sits in each network and connects them through your approved hand-off method. The design follows your network rules.
- Will it stop when the core system's screens change?
- When buttons move or fields are renamed, AI reads the screen again, relearns the step, and carries on. What it changed stays in the run record. When AI cannot fix it, the person in charge is told where it stopped.
- Could it register something by mistake?
- A person can approve before anything is registered or updated. A step can read the screen back afterwards to check it went in correctly. Every run is recorded with screenshots.
- Does the AI use outside services?
- You choose the AI model: your own Azure OpenAI, Amazon Bedrock, or Google Vertex AI contract, or an OpenAI-compatible model on your own servers. With an in-house model, screens and documents stay inside. Repeating learned steps uses no AI.
- Can it work terminal emulator (host) screens?
- It works terminal emulator screens on Windows the way a person does. We confirm it on your screens in the check.
- Can our in-house systems start Kitewell?
- In-house systems and AI agents can call it over MCP. It also starts on a schedule, when an email arrives, or for each row of an Excel list.
- What does it cost?
- The check and seeing it connected are free. The one-month joint verification is ¥500,000 before tax, with no charge if the success criteria are not met. Licenses are ¥50,000 per on-premises computer per month before tax; building and maintenance are quoted.
- How is it maintained afterwards?
- AI fixes changed screens. When a step should change, your team changes it by talking with AI. Descarty's maintenance support is also there, for example to find why a run stopped.
Get a check of the systems to connect
Tell us which hand-offs between your in-house systems people still do by hand. We tell you which door to use and how to proceed. The check is free.