Early access partners

We are looking for companies to verify Kitewell with us on their real work and carry it through to production. Kitewell is in public beta, and Descarty fixes in the product what the verification finds. The results are published as joint verification case studies, without your company name if you prefer.

Apply for early access Download the overview

Program details

What the joint verification (PoC) covers, how it runs, and what each side brings.

3 companies, until March 31, 2027

Verify Kitewell with us on your own work

Choose one to three tasks from robots that stop when screens change, integration with systems that have no API, and document work that needs judgment or approval, and we run Kitewell workflows on the same screens and data as production.

Kitewell's developers support you directly, from the first interview to production
Features your work needs are added to Kitewell itself, with priority
The results are published as a joint verification case study, without your company name if you prefer
Terms are agreed individually, once we know the work and its scale
Kitewell is in public beta (0.x); features and screens may change

Apply as an early access partner

How it runs

0
Interview 30 minutes, online
We learn about your current work, problems, and use of RPA.
1
Agree on scope and success criteria About 1 week
Choose one to three tasks and agree on success criteria such as processing time and success rate.
2
Build 2 to 3 weeks
We build the workflows on your real screens and data, together with your team.
3
Verify and report 1 to 2 weeks
We run them on real data and report the results with a proposal for adoption.

Typical length: 4 to 6 weeks in total

What you receive

The workflows run during the verification, with their knowledge
A robot ledger (Excel)
A verification report: processing time, success rate, why runs stopped and how that was handled, and the estimated effect
A proposed setup for adoption

What we ask of you

One contact person for the work being automated
Test accounts and access to the screens involved
Sample data close to production
A machine to run Kitewell
An API key for the AI model you will use, or an in-house model server
Regular feedback during the verification
Help publishing the joint verification case study

Features for companies

What a company needs to manage Kitewell. You can start with one department and a few computers.

Shared computers in your Azure, AWS, or Google Cloud, on premises, or hosted by Descarty, running with nobody signed in
A browser workspace for building and fixing workflows on any computer
The AI you contract (Azure OpenAI, Amazon Bedrock, Google Vertex AI) or in-house models
Unattended system integration: your systems call it by webhook or REST API
Single sign-on (SAML) and automatic provisioning of people (SCIM)
Audit-log export and long-term retention of every change and run on every computer
Fleet management of computers: status, version, last connection, and what runs where
Roles on each project, and policies for the sites and apps allowed
Closed-network setup that keeps sync inside your own environment
Payment by invoice (bank transfer), in yen, on an annual contract
An SLA with defined response times
Support for your security review and questionnaires
OEM and white-label supply under your own brand

OEM and embedding

Example setups

Three common setups. In the joint verification, we fit the setup to your environment.

Entirely inside your Azure

Kitewell is deployed to a VM in your Azure, and AI is Azure OpenAI in the same tenant. Business data never leaves your Azure.

Requests

SchedulesEvery morning at 8, at month end
PeopleRun from a browser with a company account

Your Azure

KitewellA Windows VM that runs with nobody signed in
Azure OpenAIModels in the same tenant
Internal systemsScreens and APIs over the virtual network

Results

Microsoft TeamsResults and approval requests
Partner websitesWorked from a fixed outbound IP

A closed network with an in-house model

It runs inside a network with no internet access, on your own server and your own model. Nothing leaves the network.

Requests

SchedulesOvernight batches and more
PeopleRun and approve from an internal browser

Your closed network

KitewellYour own server
In-house model serverOpenAI-compatible, such as vLLM
Core systemsScreens and APIs

Results

Internal email and chatResults and approval requests
Internal file serversResults as Excel or CSV

An unattended part of your systems

Your system sends a request by webhook or API, Kitewell works the screens, and the result goes back. People step in only when a decision is needed.

Requests

Your systemRequests by webhook or REST API
AI agentsRequests over MCP

Your cloud

KitewellA cloud VM that runs with nobody signed in
AI modelBedrock, Vertex AI, or Azure OpenAI
Screens it worksBusiness systems and partner sites with no API

Results

Your systemResults returned over HTTP
PeopleApprove only what needs a decision

Shared computers (enterprise option)

Keep running Kitewell on each person's PC, and add company computers, such as a cloud VM or a dedicated PC. They run with nobody signed in; people build and fix workflows from a browser, and approve a tested fix from their phone when a run stops.

An illustration. The screens are drawn to explain the option and differ from the real app.

Build support, maintenance, and integration

After the joint verification, we support you through to production. Your team builds the workflows with AI; Descarty helps choose what to automate, design the exceptions and approvals, and move the work into production.

Build support

Team and Enterprise

We help your team turn tasks into workflows one at a time, on your real screens and data.

The scope of support and what counts as done, agreed first
Choosing the task and designing its exceptions and approvals together
Help when your team gets stuck while building

Maintenance and support

Enterprise only

Maintenance of Kitewell itself, and answers while you run it. Your team fixes its workflows, and Descarty helps with how.

Answers to your questions about Kitewell
Finding why a run stopped when the AI could not fix it
Releases that fix Kitewell bugs
Checking each update against your workflows first
A monthly report: items processed, and how often and why runs stopped

Integration and features

Team and Enterprise

We design the integration with your core systems, internal tools, and products, and add the features your work needs to Kitewell itself. They become part of the product, so every later release keeps them.

Integration with core systems and internal tools, and how your systems call Kitewell
We learn your requirements and propose the scope and delivery date
Only features that fit the product; we do not make a private modified version
Features we add become available to other customers too

Hosted shared computers

Enterprise only

Descarty sets up and runs a VM dedicated to you as a shared computer, so you can start without a cloud contract or VM administration of your own.

A VM dedicated to you, never shared with other customers
OS and Kitewell updates, backups, and uptime monitoring
A fixed outbound IP you can add to a supplier site's allowlist
Descarty's operators administer the VM. If credentials must stay in your environment, put the shared computer in your own cloud instead

Ask about build support or maintenance

Frequently asked questions

What does it cost?

Terms for early access partners are agreed individually, once we know the work and its scale. You pay AI model usage fees directly to the model provider. There are none if you use in-house or local models.

Can we take part without naming our company?

Yes. A joint verification case study can be published with only your industry and size. You review what is published before it goes out.

Where does Kitewell run?

On each person's PC. Under an enterprise plan, workflows can also run on shared company computers: a VM in your own Azure, AWS, or Google Cloud, a server or dedicated PC on your premises, a closed network, or a VM Descarty hosts. Run logs, results, and credentials are stored on the computer that runs the workflow. Supported platforms are:

Platforms

macOSBeta
WindowsBeta
LinuxComing

Can we choose how Kitewell is deployed?

Yes. Under an enterprise plan you choose where workflows run to fit your requirements: each person's PC, your own Azure, AWS, or Google Cloud, on-premises, a closed network, or a VM Descarty hosts. AI can be your own Azure OpenAI, Amazon Bedrock, Google Vertex AI, or in-house models. Licensing, team sync, and updates can all stay inside your environment, and paired with an in-house or local AI model, nothing is sent outside. We settle the setup during the joint verification.

Do we need to remote-desktop into a shared computer?

Only for first-time setup, such as installing a desktop app. Building, trying, fixing, and approving workflows all work from a browser and a phone.

Can we use coding agents such as Claude Code on a shared computer?

Yes. On a shared computer, run them with API keys or the vendor's business plan. A subscription bought by one person may, under its terms, be limited to that person's own use.

Which AI models can we use, and is our data used for training?

Major AI model APIs, your company's own Azure OpenAI, Amazon Bedrock, and Google Vertex AI, and in-house or local model servers that speak the OpenAI API. Under their terms, the paid APIs of Anthropic, OpenAI, and Google Gemini, Azure OpenAI, Amazon Bedrock, and Google Vertex AI do not use what you send for training by default; Gemini's free tier does, so use a paid API for business data. Through OpenRouter, the assistant's requests go only to providers that neither store nor train on them. With in-house or local models, data is not sent outside, though small local models usually cannot operate a screen. Descarty never receives run results or logs, and never uses synced workflows or knowledge to train AI or shares them with other customers. Switching models keeps your knowledge working. See "Is your data used to train AI?" on the security page for details.

What happens to our automation when the person in charge moves on?

As you work, the assistant writes down in each workflow's knowledge what it is for, why it was built that way, and how it was fixed when it broke, and reads it before it answers. Whoever takes over reads the knowledge and asks the assistant, and carries on running it. A robot ledger also exports to Excel from the definitions and knowledge.

Where is our data stored?

Workflows, knowledge, run logs, and credentials are stored on the computer running Kitewell. When a project is synced through Dagu Cloud, its definitions and knowledge are stored in a Google Cloud region in the United States; secret values, run results, and logs are never sent. Requests received by a webhook wait at the relay for at most 7 days until the computer takes them. When people use a shared computer from a browser, logs and results still stay on that computer, and passwords typed in the browser are encrypted for that computer before they are sent. Approval inputs and send-back notes pass through Dagu Cloud. Enterprise plans can keep sync within your own cloud or on-premises.

Do we have to replace all of our RPA?

No. We recommend moving the work that breaks often and matters most, and leaving work RPA handles well, such as fixed operations replayed at high speed, where it is.

Can it automate desktop applications?

Yes. Besides web browsers, Kitewell operates desktop applications on macOS and Windows the way a person does: AI reads the screen and uses the mouse and keyboard. Commands, scripts, APIs, and server operations can be built into workflows as steps.

Do you support single sign-on (SSO)?

Yes, under an enterprise plan: single sign-on through SAML, and SCIM to add and remove people automatically. Personal and Team plans sign in with a Dagu Cloud account.

Can we pay by invoice?

Yes. Enterprise plans can be paid by invoice (bank transfer), in yen, on an annual contract.

Who is the contracting party, and in which languages do you work?

Your contract is with Descarty, Inc., headquartered in Tokyo. We work in Japanese and English.

Apply as an early access partner

Tell us which work you have in mind and where things stand. We usually reply the same business day.

Apply for early access Download the overview