OPUS Agent Manager overview

OPUS Agent Manager enables users to define, create, release, and administer agents in OPUS. Agents help perform business work by using a defined business intent, measurable objectives, instructions, and guardrails. Users first define a reusable agent profile, then create one or more agent instances from that profile.

An agent profile describes what an agent is expected to do, why it is useful, how it should work, and where it is allowed to operate. After the profile is released, an Agent Administrators can use it to create an agent instance for a specific company, process network, or enterprise service.

Why use OPUS Agent Manager?

OPUS Agent Manager helps organizations create agents in a controlled and outcome-driven way. Instead of defining agents only as automation tasks, OPUS Agent Manager focuses on business outcomes. This helps users create agents that are easier to understand, test, release, and administer.

For example, in the MINT Purchase Orders process, users can create, submit, search, view, reprocess, and resubmit purchase orders exchanged with Partners. An Agent Developer can define an agent profile that helps monitor purchase orders that need attention, such as draft purchase orders, recently modified purchase orders, or purchase orders that may need to be reprocessed or resubmitted. The profile can define the agent’s intent, expected outcomes, success metrics, instructions, and guardrails. After the profile is released, an Agent Administrators can create an agent instance for the relevant company and process network. The agent instance can help the assigned human manager quickly find purchase orders that need review and summarize the next action. This reduces manual searching and helps users focus on reviewing, correcting, submitting, or resubmitting purchase orders instead of scanning order records manually.

With OPUS Agent Manager, users can:

Define reusable agent profiles that describe the purpose, goals, operating instructions, and boundaries of an agent.
Test agent behavior before the profile is released for use.
Release agent profiles to a catalog so they can be reused.
Create company-specific agent instances from released profiles.
Assign a human manager or human user to an agent instance, depending on the agent type.
Administer agent instances by evaluating, activating, cloning, retiring, and viewing performance information.

Before using OPUS Agent Manager, it is important to understand the difference between an agent profile and an agent instance.

Agent profile

An agent profile captures the business purpose of the agent and the rules that guide how the agent should operate. Agent profiles are created and maintained before agent instances are created for actual use.

An agent profile typically includes:

Intent - Defines the primary business function the agent performs.
Objectives - Define the specific outcomes the agent should accomplish.
Metrics - Define how success is measured.
How-to instructions - Define how the agent should work in a business context.
Guardrails - Define the boundaries and limits for the agent.

After an agent profile is complete and tested, it can be released to the catalog. Released profiles become reusable items that Agent Administrators can use to create agent instances.

Agent instance

An agent instance is a company-specific agent created from a released agent profile. The agent instance inherits its definition from the profile, but some details can be configured for the specific company, process network, enterprise service, or business context in which the agent instance will operate.

For example, when creating an agent instance, an Agent Administrators may provide a name, select the relevant process network or enterprise service, and assign a human manager or human user, depending on the agent type.

After an agent instance is created, it follows its own lifecycle. It can be evaluated, moved to In Service, Retired, or Cloned.

How agent profiles and agent instances work together

Agent profiles and agent instances work together in a two-part model. The profile defines the reusable design of the agent. The instance applies that design to a specific operating context.

This model helps organizations reuse a tested agent definition while still allowing controlled configuration at the instance level.

An Agent Developer creates an agent profile.
The Agent Developer defines the profile intent, objectives, metrics, how-to instructions, and guardrails.
The profile is tested by creating agent instances in a supported test environment.
The profile is released to a catalog.
An Agent Administrators browses the catalog and creates an agent instance from the released profile.
The Agent Administrators configures the agent instance for the company, process network, enterprise service, or human user.
The agent instance is evaluated and then moved to In Service when it is ready for use.

Outcome-driven agent definition

OPUS Agent Manager uses an outcome-driven approach to define agents. This means that the agent is defined around business results rather than only around tasks or automation steps.

The outcome-driven model can be understood using four questions:

Question	Agent profile part	Description
What should the agent do?	Intent	Defines the primary business function the agent performs.
Why does the agent exist?	Objectives and metrics	Defines the business outcomes the agent should achieve and how success is measured.
How should the agent work?	How-to instructions	Defines the business objects, operations, and instructions the agent uses to achieve its objectives.
Where can the agent operate?	Guardrails	Defines the scope boundaries, data boundaries, and operation limits that control what the agent can and cannot do.

Agent instance lifecycle

An agent instance is created from a released agent profile and then administered through its own lifecycle. The instance lifecycle helps Agent Administrators evaluate, activate, and retire the agent instance.

The agent instance lifecycle is:

Evaluate: The Agent Administrators reviews and evaluates the agent instance before it is used for active work.
In Service: The Agent Administrators moves the agent instance to In Service when it is ready to perform work in its assigned context.
Retired: The Agent Administrators retires the agent instance when it should no longer be used.

Agent Administrators can also clone agent instances when they need to create a similar instance with a different human manager or assigned membership.

Agent types

OPUS Agent Manager supports different agent types. Each type represents a different level of autonomy and human involvement.

Managed Agent

A Managed Agent is a virtual worker with its own identity and permission set. A managed agent has an assigned job role, also referred to as its intent, and one or more objectives that define the outcomes it should achieve.

Managed agents can perform work in OPUS, but their actions are supervised or reviewed by a human within the organization. This provides oversight and accountability while still allowing the agent to support business work.

Use a managed agent when the agent should perform work like a digital resource, but human supervision is still required.

Actor

An Actor is an autonomous agent that can perform tasks and make decisions without human supervision or approval. Actors share some structural characteristics with managed agents, such as intent and objectives, but they are designed for scenarios where continuous human oversight is not required.

Use an actor when the agent must operate independently and complete tasks without direct human approval.

Assistant

An Assistant supports and augments human work. Assistants do not act independently or make autonomous decisions. Instead, they operate within the context of a human user's permissions and help by providing suggestions, completing sub-tasks, or offering relevant information.

Use an assistant when the agent should help a human user perform work rather than perform the work independently.

Roles and responsibilities

Different users perform different tasks in OPUS Agent Manager.

Role	Responsibilities
Agent Developer	Creates, defines, tests, and releases agent profiles. Agent Developers define the intent, objectives, metrics, how-to instructions, and guardrails that determine how an agent should work.
Agent Administrators	Creates agent instances from released profiles, configures instance-specific details, evaluates agent instances, moves instances to In Service, clones instances, retires instances, and views performance information.
Human Manager	Supervises or reviews the work performed by a managed agent. A human manager provides accountability for agent activity within the organization.

What can be configured on an agent instance?

When an agent instance is created from a profile, some parts of the inherited profile definition can be adjusted for the instance. Other parts remain locked to protect the outcome design of the agent.

Profile part	Can be changed on the agent instance?
Intent	No. The intent cannot be changed on the agent instance.
Objectives	Partially. Existing objectives can be removed, but new objectives cannot be added.
Objective goals	No. The goal of an objective that remains on the instance cannot be changed.
Objective metrics	Yes. Metrics can be changed for the agent instance.
How-to instructions	Yes. How-to instructions can be updated for the business context of the instance.
Guardrails	Yes. Scope boundaries, data boundaries, and operation limits can be modified for the instance.

Catalogs and released profiles

Released agent profiles are stored in a catalog so they can be reused to create agent instances. The catalog makes profiles available to Agent Administrators based on the type and scope of the profile.

Agent profiles can be available as:

Standard profiles, which are released to the global catalog.
Company profiles, which are released to the company catalog.

What happens when an agent instance is created?

When an Agent Administrators creates an agent instance, the Agent Administrators provides the details needed to make the instance specific to the company and operating context.

The Agent Administrators may provide:

A name for the agent instance.
Allowed changes to the profile details inherited by the instance.
A process network or enterprise service, depending on where the agent will operate.
A human manager for a managed agent, or a human user for an assistant.

OPUS Agent Manager also performs required setup steps automatically. For example, it can create an integration principal for the agent instance and associate the required membership with the selected process network or enterprise service.