Copilot Studio Integration and Nisus Central
Copilot Studio talks.
Central runs the operation.
Status: The Nisus Central package is in early access. Today the connection is set up through Central Agent's local HTTP interface, and the NisusAI team sets it up with you.
Nisus Central connects AI agents built with Copilot Studio to the company's robots and approval processes. The agent writes the request to a queue, has a robot do the work and waits for an employee's approval on critical decisions; every step is recorded and each approval is kept with the user's identity. Central is installed on premises, so the data stays on the company's servers.
The need
What AI agents need at enterprise scale
Building an agent in Copilot Studio is fast. Once agents enter business processes, governance questions grow:
- Seeing which robot completed the work an agent started, and when
- Human approval for AI decisions, recorded with user identity
- Agent and robot errors in one place, by category
- Central control of the credentials an agent can use
- FTE reporting for the savings AI agents deliver
- Process data that stays on-premises
Copilot Studio licensing and capacity terms still apply. Central does not change the agent; it adds the management layer.
Integration
Two ways to connect a Copilot Studio agent to Central
The agent runs in the cloud; Central runs on your infrastructure. That gives two paths: a desktop flow bridge, or a secure API call to Central services.
Set up the Agent and robot
Install the Nisus Central Agent on the machine where the work runs and enter the robot key.
Path 1: desktop flow
Copilot Studio triggers a Power Automate desktop flow; the flow adds work through the Agent's local bridge and returns the result.
Path 2: API call
The agent calls Central services with an HTTP request or custom connector, using a robot identity.
Monitor and approve
Work started by the agent, approvals and errors appear on the Central dashboard with your robots.
Nisus Central
What Nisus Central adds to Copilot Studio
Queues and work distribution
Copilot Studio picks work items from Central queues. Status, retry count, deadline and postpone date live in one place.
One screen for every platform
Copilot Studio on the same dashboard as UiPath, Blue Prism, Power Automate, Automation Anywhere and n8n.
Human approval
The flow pauses at an approval step. The decision is recorded with the user's identity and the work resumes where it stopped.
Central credentials
Usernames and passwords are stored in Central. The flow retrieves them securely at run time.
Error records
Error category, error detail and a screenshot are kept for every run.
Machine monitoring
Online status, CPU, memory and disk usage of every machine, live.
Scheduling
Cron-based schedules, dynamic robot allocation and a concurrent-job limit.
FTE and cost reporting
Hours returned to the business and savings, ready for management.
Ready functions
Which path, when?
Use one path or combine several, depending on the need.
| Capability | How it works |
|---|---|
| Add work to a queue | The agent writes the user's request to a queue; robots process it in turn |
| Start a job | The agent starts a defined process on a robot and checks its status |
| Human approval | The agent's proposed decision goes to approval in Central, and the result returns to the agent |
| Status questions | A user asks "where is my request?" and the agent answers from the queue |
| Analytics questions | The Nisus Central MCP server answers process, robot, transaction summary and FTE questions within user permissions |
| Security | Robot identity, role-based permissions and an audit record for every call |
Comparison
Copilot Studio alone vs. Copilot Studio + Central
| Topic | Copilot Studio alone | Copilot Studio + Central |
|---|---|---|
| Conversation and agent design | Yes | Yes, stays in Copilot Studio |
| Same queue as robots | No | Yes |
| Human approval with identity | Limited | Yes, exportable |
| Agent and robot errors in one view | No | Yes |
| Central credentials | Per environment | Central |
| FTE and cost reporting | No | Yes |
| Process data location | Microsoft cloud | On-premises, your own servers |
Example scenario
The agent talks, the robot delivers
An employee request moves along an iterative line shared by an AI agent and robots.
- 1
Request
An employee sends a leave or expense request to the agent in Teams.
- 2
Queue
The agent adds the request to a Central queue.
- 3
Process
A robot takes the next request and enters it into the HR or ERP system.
- 4
Approve
Out-of-policy requests go to the manager for approval.
- 5
Inform
The agent reads the status from the queue and answers the employee.
- 6
Report
Request volume, duration and hours saved on the dashboard.
Fit
Who is it for?
A good fit
- Organisations putting AI agents into processes alongside RPA robots
- Teams that want human approval and an audit trail for AI decisions
- Leadership that wants agent and robot performance in one report
- Sectors that must keep process data on-premises
Not a fit
- Q&A-only agents that are not connected to business processes
- Teams that are happy with a single platform and have no on-premises data requirement
FAQ
Frequently asked questions
How are AI agents used safely in enterprise processes?
With three rules: the agent stops at a decision it is unsure about and asks an employee, every action it takes goes into the audit trail, and the agent sees only the data it is authorized for.Nisus Central applies these three rules to Copilot Studio agents.
What happens if an AI agent makes a wrong decision?
Critical decisions are not left to the agent.The agent sends the decision to approval in Central, and nothing is done until the approval arrives. If something goes wrong, the record shows who approved what and when.
How are an AI agent's actions audited?
Every job the agent starts is a record in the Central queue: who asked for it, which robot did it, who approved it and what the result was.The records can be exported.
How does a Copilot Studio agent run robots?
By adding work to a queue or starting a defined process.Robots take the work from Central and write the result; the agent can check the status.
How does a cloud agent reach an on-premises Central?
Either through a desktop flow on a machine with the Agent, or through a Central service address your organisation exposes securely, using a robot identity.
Can Copilot Studio connect to Nisus Central over MCP?
The Nisus Central MCP server reads process, robot, transaction and FTE data within each user's permissions.It suits analytics questions.
Can we add human approval to AI agent decisions?
Yes.The decision goes to approval in Central and is recorded with the approver's identity.
How is governance handled for Copilot Studio agents?
Jobs, approvals and errors started by the agent are kept in Central with a robot identity, role-based permissions and an audit record for every call.The credentials the agent can reach are managed centrally in Central too.
How does a request made to a Copilot Studio agent in Teams reach a robot?
The agent adds the request to a Central queue; a robot takes the next request and enters it into the HR or ERP system.Requests outside the rules go to the manager for approval, and the agent reads the status from the queue to answer the employee.
What should we consider before using the API path?
For an agent running in the cloud to reach Central services, your organisation needs to expose that address securely.Organisations that prefer not to do this use the desktop flow path, which needs no publicly reachable address.
Run your AI agents and robots from one place.
Let's build one agent scenario with your own data.