Build it with your own team
Your team creates and maintains the agent and the UiPath workflow that calls it. Choose your own runtime and connect the workflow through REST.
AI workflows / for existing UiPath teams
Add an agent.
Keep UiPath in control.
Build your own independent agent, or have TaskSultan configure the agent and matching UiPath workflow for you. TaskSultan-built packages are available from your private customer portal. Run the worker in your local environment and call it from a standard UiPath workflow through REST.
UiPath continues to run the desktop or browser automation, access local files and decide what happens next. The agent returns structured output for the workflow to validate and use.
Customer environment / local execution
UiPath Robot
Runs local apps, files and the business workflow.
REST
request / result
request / result
Agent Worker
Customer-built or TaskSultan-built, running on an approved customer machine or server.
TaskSultan-built agent and matching UiPath workflow packages are delivered through your private customer portal.
Prepare the quote.
Keep a person in control.
A customer needs laptops for 12 remote employees who use Excel, Teams and Power BI, travel often, and have a RM45,000 budget. The agent weighs those needs against product specifications and sales guidance, then explains its recommendation.
The agent handles the product-fit judgement. UiPath handles the repeatable work around it: data lookup, calculations, document creation, approval routing and email.
Illustrative scenario with fictional customer and catalogue data—not a customer case or live agent run.
Plan an AI agent integrationExample / sales quotationUiPath → local agent → UiPath
Example customer request
“12 laptops for remote staff using Excel, Teams and Power BI. They travel frequently. Budget: RM45,000.”
TaskSultan agent / recommendation
Model A · 16 GB · lightweight
Recommended from the fictional catalogue for frequent travel and the requested workload. The agent returns the reasons and catalogue references for UiPath to check.
UiPath / quote draft
Word or PDF
UiPath applies catalogue prices, agreed discounts and tax, then prepares the customer-ready draft.
UiPath prepares a draft from the checked recommendation.
Required human approval / before send
Review the quote and approve or return it
The reviewer checks the recommendation, quantities and final quote. Only an approved quote reaches the UiPath send step.
01
Robot
UiPath reads the request
Read the customer enquiry from Outlook and the sales tracker; retrieve catalogue prices and customer terms from Excel.
02
AI agent
TaskSultan agent recommends
Search product specifications and approved sales guidance. Return model IDs, reasons and any unresolved assumptions as structured data.
03
Robot
UiPath builds the quote
Validate the recommendation, calculate prices, discounts, tax and totals, then generate a Word or PDF draft.
04
Human approval
A person reviews before sending
A mandatory approval task presents the quote for review. Approve it to continue, or return it for changes. Nothing is emailed to the customer before approval.
05
Robot
UiPath sends after approval
On approval, email the quote to the customer and update the sales tracker. If it is rejected or returned, keep it out of the send path.
Choose who builds the agent.
Choose where the model runs.
These are two separate decisions. A locally running agent worker can call a cloud model, or connect to a model hosted on your own infrastructure.
Agent build / ownership
Have TaskSultan build it
TaskSultan configures the agent and matching UiPath workflow, then prepares both for download from your private customer portal.
Model hosting / inference
Cloud model
The agent worker calls a model API. Agree which data the workflow sends, which provider receives it and any usage charges before deployment.
Self-hosted model
Run the model on customer-managed infrastructure. This needs additional hardware sized for the model and workload, plus someone to operate and maintain it.
A local worker does not automatically mean the model runs locally. The agent location, model endpoint, data boundary, hardware and operating responsibilities are agreed before deployment.
TaskSultan-built path / implementation and handover
From agent setup to a local REST call.
When you choose TaskSultan to build it, we define the agent contract to fit the UiPath workflow that already runs your process, then package the agent and workflow together for portal download.
We configure the agent
Define its task, approved model, tools, business knowledge and response format.
Download both packages from your portal
TaskSultan publishes the configured, versioned agent package and its matching UiPath workflow to your private customer portal.
Run it in your environment
Deploy the agent worker on the approved customer PC or server. We document the runtime and configuration it needs.
Call it from UiPath
Your local UiPath workflow sends a REST request, checks the structured response and continues its normal process.
Clear responsibilities / customer-controlled
The Robot acts.
The agent assists.
The agent does not take over the automation. UiPath supplies the approved context, receives a defined response, then applies the business rules and continues or routes the case for review.
A locally running worker does not by itself mean model inference stays local. We agree the model endpoint, data sent, credentials and any hosting or API costs before deployment.
Implementation principles
- UiPath remains responsible for local applications, files and business execution.
- The agent handles a bounded reasoning task and returns a result for the workflow to validate.
- The worker runs locally; whether the selected model also runs locally or uses a model API is agreed during design.
- Approvals, retries, logging and exception handling are designed around the process and its risk.
UiPath also supports invoking its own published agents from Studio workflows. TaskSultan is for customers seeking an independent, locally deployed worker and a separately configured model, tools or business-knowledge connection. Maestro remains an option when the whole process needs end-to-end orchestration.
Ready to add an agent to a real process?
Start with one workflow.
We’ll review the process, data boundary, local environment and model requirements before agreeing the agent package and integration.