Turn robots that breakinto workflowsyou fix by talking.

AI handles the web and desktop operations, people make the decisions. Business automation you build and fix by talking to an AI assistant. The reason for every fix stays in its knowledge, so the automation keeps running when people change.

Discuss a PoC Download the overview

Platforms

macOS
Windows
LinuxComing

A Kitewell run: a workflow that collected orders from a supplier portal is waiting for the owner's approval — Actual screen: a workflow waiting for approval

The interface comes in six languages, including Japanese and English. Chat and steps work in any language your chosen AI model supports. Screens are from a demo environment with fictional data.

When your automation stops, can you fix it?

It stops when a screen changes
A small change to a web system stops the robot.
With KitewellIt repeats the remembered steps the same way every time, and only when the screen changes does the AI relearn them and carry on.
Only the builder can fix it
When that person moves on, nobody can.
With KitewellAs you work, the assistant writes down what each workflow is for and why it was fixed.
Nobody keeps track
Robots multiply across departments, unseen.
With KitewellEvery run and change is recorded, and a robot ledger exports to Excel.

About moving from RPA

What AI can operate

Everything RPA could not reach, in the same workflow.

Web systems
Supplier portals, SaaS, and internal web systems. Write the steps in words, and AI operates Chrome the way a person does. A small screen change does not stop it.
Sign-ins stay on that machine. Passwords are never shown to the AI.
Feature details
Desktop applications
Sales and accounting software and dedicated clients with no API. On macOS and Windows, AI reads the screen and works the mouse and keyboard.
It waits while a person is using the mouse or keyboard, and an approval can sit before a save.
Feature details
Email
Gmail, Microsoft 365, and any IMAP mailbox. Find what arrived, have AI read it, record it in a ticket or ledger, and organize what was handled.
Connections and credentials stay on the machine. A failed email is handled again next run.
Feature details
APIs, servers, and scripts
Imported OpenAPI operations, servers over SSH, commands and scripts, Docker. They sit in the same workflow as the screen operations.
Credentials are encrypted secrets, never written into the workflow.
Feature details

Build and fix in the language of the work.

Ask, and the workflow is built

Say what you want in your own words, and the assistant builds a workflow and proposes it.

Nothing is saved until you choose Apply.

Feature details

A workflow the assistant proposed, with Edit in editor, Reject, and Apply buttons — Actual screen: nothing the assistant proposes is saved until you choose Apply

Screen operations are sentences

Write steps such as “enter the login ID”, and AI reads the screen to carry them out, in a browser or a desktop application. An operation that worked is remembered and repeated the same way on later runs, without the AI. A small screen change does not mean rewriting the steps.

Passwords in web steps are named, never shown to the AI.

Feature details

The workflow editor, with web steps written in plain language such as entering the login ID — Actual screen: web steps written in plain language

When it stops, fix it on the spot

From the step that stopped, ask the assistant about the cause. Once fixed, run again from that step.

Completed steps keep their results.

Feature details

A Kitewell run: recording to the sales system failed, with an Ask the assistant button — Actual screen: the step that stopped, and a button to ask the assistant about it

What you teach becomes the team's knowledge

Tell the assistant a fact, a rule, or a fix, and it writes it into the knowledge, then reads it before it answers, whether you build a new workflow or fix one. Whoever asks starts from the same place, so the more you use it, the less anyone has to explain.

The lines it changed are marked and can be undone at once. After reading a web page or a run's output, it asks before saving.

Feature details

A knowledge page the assistant just updated: the line it added about a one-time code is marked, with Undo and Keep, beside the assistant's receipt and reply — Actual screen: the line the assistant added is marked, and can be undone at once

Decide what AI does and what people check

A batch sheet listing rent, fees, and status for eight properties, with the quote behind one rent value — Actual screen: values AI read, with the quote behind each

Run on each PC. Share every fix.

The whole team works from the same workflows and knowledge, with no management server to build.

Runs with each person's sign-ins
Workflows run on each person's PC or server with their own sign-ins and credentials. No shared robot account is needed.
Every fix reaches every device
A workflow or page someone fixes reaches the team's devices on its own. On the paid plans (Personal and Team), even an unattended device picks up the new version about every 5 minutes.
Roles for each project
Give each person a role on each project. Each run is recorded on the machine that ran it.

Fixes are shared only within your team. Descarty never uses your workflows or knowledge to train AI, and never shares them with other customers.

About team sync

Your business data stays in your environment

Workflows, knowledge, run logs, and credentials stay on your machine
You choose the AI model, including in-house and local models, and switching keeps your knowledge
Credentials are encrypted, and AI never sees their values
Descarty never uses your data to train AI

About security