Custom Agents extend the AI Assistant beyond ad-hoc questions and answers. A Custom Agent runs a predefined, multi-step procedure on your behalf. It uses tools to gather data from your Ardoq graph and metamodel, apply built-in governance rules, and produces a structured report, recommendation or updates to Ardoq data. Each agent is purpose-built for a specific enterprise architecture task, so you get consistent, repeatable results every time you run it.
In the first release you will be able to configure and run Custom Agents provided by Ardoq. Coming soon you'll be able to create your own.
Availability: [Live now]. Requires Ardoq AI features to be enabled for your organization. Custom Agents are delivered as a bundle. To enable them, contact Ardoq Technical Support and request that the Custom Agents bundle be installed. Self-service installation coming soon
What Are Custom Agents?
A Custom Agent is a pre-configured AI assistant instruction with a fixed task, a defined set of tools, and built-in governance rules. Instead of composing a question, you invoke the agent by name and it executes a structured workflow against your Ardoq data.
Purpose-built. Each agent targets one EA task - for example, rationalizing an application or mapping stakeholders.
Multi-step and structured. Agents follow a defined sequence of instructions and return a consistent report, not a one-off answer.
Grounded in your metamodel. Agents read your metamodel definitions before drawing conclusions, so interpretation respects how your organization has modeled its architecture.
Governed. Agents follow rules, such as respecting decision-process sequencing, treating calculated fields as derived values, and using reference directionality, so outputs stay trustworthy.
Permission-aware. Like the assistant, an agent only sees data the user is authorized to see.
Getting Started
Simply list the custom agents and then run a custom agent
The agent will ask you to fill in necessary details
list the custom agents
Run the Component Overview agent
Agents Available at Launch
The initial launch set of agents in the Custom Agents bundle includes a set various agents which will be expanded over time. For instance
Agent | What it does | Prerequisites |
Component Overview Agent | Summarizes what a component is and how it is used across the business, combining web search* with the component’s connections in your graph. Produces a concise, fully source-attributed report.
(* web search is not enabled yet but will roll out during Q3) | None |
Application Rationalization Agent | Analyzes a single application across attributes, business context, technology products, contracts, initiatives, decision records, solution health checks, hosting infrastructure, risk, and technical debt to support a rationalization decision — and flags missing information. | Application Rationalization Solution (or equivalent viewpoints) |
Stakeholder Map Agent | Identifies the stakeholders of a component, recommends Interest and Power scores with justifications, and plots each onto a stakeholder-map quadrant. Can optionally write stakeholder references and scores back to Ardoq. | Stakeholder-type components (eg. People or Roles) in the metamodel; Interest/Power fields for optional write-back |
Use the Contract Extraction Agent to automatically analyze a software contract document and populate Ardoq with the supplier, application, technology product, owners, and organizational consumers. | SW Contracts, Technical Product Management, and Foundation | |
Disaster Recovery Blast Radius Agent | The DR Blast Radius Agent produces a structured disaster recovery blast radius analysis for a given application. It answers the question: "If this application fails, what is affected and how ready are we to recover?" | |
Various Foundation Insights Agents | A set of agents will replace the Foundation Insights Agent. These will provided a more flexible version of the same functionality that isn't directly dependent on the Foundation Metamodel | None |
To get the full list of custom agents available in your Ardoq organization run the following command in the AI Assistant:
list the custom agents
You can follow up and ask the AI Assistant to run any of the agents in the list
Custom Agent Help
Ask the Assistant how a particular Custom Agent works and how to run it.
How a Custom Agent Is Built
Every agent is defined by a small set of properties. Knowing them helps you understand what an agent will do before you run it.
Property | What it defines |
Name | How you refer to the agent when you run it. |
Description | What the agent does, its prerequisites, and example commands for running it. |
Instructions | The step-by-step procedure the agent follows, including its governance rules. |
Tools | The Ardoq AI Assistant tools the agent may use — e.g., searching components, reading the metamodel, running viewpoints, and (for some agents) writing data back. |
Variables | Placeholders that let users configure the functionality of the agent for their own metamodel - component types, viewpoints to use, workspaces, and other metamodel type. |
Model & reasoning | Whether the agent uses extended thinking and a more advanced model for complex analysis. |
Flexibility with Variables
Custom Agents can have variables — written as <<variable_name>> — so a single agent can be reused across different components, workspaces, and metamodels without being rewritten.
Sensible defaults. Many variables have defaults. The Component Overview Agent, for example, defaults <<component-type>> to Application and <<component-workspace>> to Applications, so you can run it without specifying them.
Override on demand. Supply your own values in the command to point the agent elsewhere — for instance, run the Component Overview Agent against a Business Capability instead of an Application.
Metamodel-aware resolution. Agents resolve variables against your metamodel. If a value cannot be resolved, the agent lists what is missing and asks you to clarify rather than guessing.
Graceful fallbacks. Agents check whether a component type exists in your metamodel before using it, and skip or adapt steps when it does not — which is what lets the same agent run across differently-modeled organizations.
Example:
execute the Component Overview Agent for the Payments application in the ‘B2C Applications’ workspace.
Solution Dependencies
Some agents are designed to work with a specific Ardoq Solution and its bundled assets(ie. Viewpoints, Reports or Dashboards); others are metamodel- and asset agnostic. Those designed for a particular Solution can be configured using the variables to match your own setup. For example:
Solution-dependent. The Application Rationalization Agent assumes you are using the Application Rationalization Solution and its associated viewpoints. If you use your own metamodel and assets for Application Rationalization, you can provide equivalent Viewpoints as alternatives.
Metamodel-agnostic. The Stakeholder Map and Component Overview agents adapt to your metamodel and do not require a specific Solution — they check which component and reference types exist before using them.
What this means for you. Before running a solution-dependent agent, make sure the relevant Solution (or equivalent assets) is in place. Agents that cannot find an expected asset (ie. a Viewpoint) will tell you, and often recommend actions before it will come with complete results.
How to Get Custom Agents
Custom Agents are delivered to your organization as a bundle.
Confirm that Ardoq AI features are enabled for your organization.
Contact Ardoq Technical Support and request that the Custom Agents bundle be installed for your organization.
Once installed, the agents are available to run from the AI Assistant for any user with access to the underlying data.
Self-service of custom agent installation and upgrades is coming soon
Example Walkthrough
Running the Component Overview Agent
Suppose you want a quick, source-attributed briefing on how Salesforce is used in your organization.
Give the AI Assistant the following instruction:
execute the Component Overview Agent for SalesforceThe agent resolves its variables — using the default type Application and workspace Applications — and confirms a component named Salesforce exists. If it cannot find one, it stops and tells you.
It reads the metamodel definitions for the component type and the connected reference types, so it interprets relationships correctly.
Because Salesforce is a commercial product, it runs a web search* for vendor and product-category context, clearly labeled as web-sourced. (*web search is coming soon)
It inspects directly connected components for how Salesforce is used internally, and runs a matching viewpoint if one applies.
It returns a concise (max ~500-word) report with fixed sections — name, description, summary of fields, how it is used in the company, and who to contact — with every claim attributed to an Ardoq field, an Ardoq reference, a viewpoint, or web search.
You can then ask the AI Assistant follow up questions as the agent's output is available in the Assistant's context.
The result is a consistent, source-attributed overview you can regenerate any time the underlying data changes.
Coming Soon: User-Defined Custom Agents
Today’s agents are built and maintained by Ardoq. User-defined Custom Agents are coming soon: you will be able to author your own agents. That includes defining a name, description, instructions, the tools the agent can use, and its variables. Then run them from the AI Assistant just like the built-in ones. This lets your organization codify its own repeatable EA workflows. [Coming Q3 2026]
Tips for Great Results
Name the agent and the target. “execute the [Agent] for [Component]” gets you started fastest.
Set up prerequisites first. For solution-dependent agents, confirm the relevant Solution or equivalent assets (ie. Viewpoints), types, and workspaces are in place.
Let the agent ask. If something is ambiguous, the agent will ask you to clarify — answer precisely for the best results.
Use follow-ups. After an agent finishes, ask follow-up questions to drill into any section of its report.
Verify before acting. Agents accelerate analysis but can make mistakes — check critical findings against the source data.
Known Limitations
AI can make mistakes. Agents use a large language model and can produce errors on complex, multi-step analysis. Always verify important outputs. Although these agents have been rigorously evaluated on multiple Ardoq instances representing different metamodels and population sizes, they can still make mistakes.
Depends on your data. An agent can only analyze what exists in Ardoq. Missing fields, references, or assets (ie. Viewpoints) reduce the quality of the analysis. Many agents will explicitly flag what is missing.
Solution assumptions. Solution-dependent agents work best when the expected model is in place; results may vary on heavily customized metamodels.
Large datasets. Very large graphs or deep-spanning viewpoints may produce slower or less reliable responses.
Security and Privacy
Permission-aware. Agents only access data the user is authorized to see.
Some agents can write data. A few agents can create and update Ardoq content. For example, the Stakeholder Map Agent can add stakeholder references and scores. All agents are instructed to ask for confirmation before updating information. All updated information is added to a scenario for manual merging to the repository.
Built into Ardoq. Agents run natively on Ardoq AI. For general detail, see the Ardoq AI Privacy FAQ.
Frequently Asked Questions
Q: How do I run a Custom Agent?
A: Open the AI Assistant and type “execute the [Agent Name]”, optionally naming the component or workspace.
Q: How do I get a list of all Custom Agents?
A: Open the AI Assistant and type “list custom agents".
Q: How do I get Custom Agents for my organization?
A: Confirm Ardoq AI is enabled, then contact Ardoq Technical Support and ask for the Custom Agents bundle to be installed.
Q: Do agents change my data?
A: Most only read and analyze. A few can write back, such as the Stakeholder Map Agent, and only after you confirm.
Q: Can I build my own agent?
A: Not yet. User-defined Custom Agents are coming soon. For now, agents are built and maintained by Ardoq.
Q: What if an agent can’t find a component or viewpoint?
A: It tells you what is missing and either asks you to clarify or recommends completing the analysis before proceeding.
Q: Which model do agents use?
A: Agents that perform complex analysis use extended thinking and a more advanced model; simpler agents use a standard configuration.
Q: Can I give feedback or request another type of Custom Agent?
A: Yes, go to the product portal and provide feedback on specific agents or request a new one