Agents are the core AI components in DuploCloud AI Suite. Each Agent is responsible for interpreting user inputs, deciding which Tools to invoke, and orchestrating intelligent responses using integrated data and logic. They serve as the execution layer for AI-powered workflows and can be tailored to a wide range of use cases, from conversational interfaces to backend automation.
The typical workflow for using Agents involves: creating the Agent in AI Studio, deploying it with configurable resource limits and replicas, and registering it for use within your infrastructure. DuploCloud supports two types of Agents: Prebuilt and Dynamic. Each offers a different development path, depending on whether you're deploying an existing containerized service or creating a prompt-driven AI workflow within DuploCloud.
Prebuilt Agents use a pre-existing container image that defines its functionality. To read about the Prebuilt Agents included out-of-the-box, see the .
To create a Prebuilt Agent, follow these steps:
Navigate to AI Suite → Studio → Agents.
Click Add. The Add Agent Definition pane displays.
Complete the following fields:
Click Submit to create the Agent. You can view your Agents on the Agents tab.
Once the Agent has been successfully completed, you can deploy and register the Agent so it can be used with .
Dynamic Agents are configured using flexible, user-defined parameters, including Tools, prompt behavior, and optional custom build variables. To create a Dynamic Agent, first create an Agent definition, and then build an Agent image based on its configuration.
Navigate to AI Suite → Studio → Agents.
Click Add. The Add Agent Definition pane displays.
Complete the following fields:
Click Submit. Once the Agent creation is complete, package your configuration into a deployable container image.
Now that the Agent is created, trigger a build to package your dynamic Agent’s configuration, including prompts, Tool selections, and variables, into a deployable container image.
Go to the Builds tab on the Agent’s details page.
Click Trigger. The Trigger Build pane displays.
Complete the following fields:
Click Submit to begin packaging your configured Agent into a runnable container image. After the build is complete, you can proceed to deploy the image and register the Agent.
Once an Agent image is created, it must be deployed. Deploying the Agent makes it available for use on your infrastructure.
Select the Images tab on the Agent page (AI Suite → Studio → Agents → select the Agent name).
Click the menu icon () next to the Agent image and select Deploy. The Deploy Image pane displays with name and image fields prepopulated.
Choose a deployment method:
Once an Agent has been successfully deployed, it must be registered so that the DuploCloud AI HelpDesk can route queries to it.
Select the Register tab on the Agent page (AI Suite → Studio → Agents → select the Agent name).
Click Register. The Register Agent pane displays.
Complete the following fields:
Click Submit to register the Agent. This Agent can now be utilized by the DuploCloud HelpDesk. To learn how to configure and use HelpDesk, see the .
You can customize the prompts that appear when interacting with an Agent by adding metadata to its configuration.
To add custom prompts for an Agent:
Go to AI Suite → AI Studio → Agents and select the Agent from the Name column.
Open the Metadata tab.
If a prompt_suggestions key exists, click the menu icon () and select Edit to update the prompt suggestions.
Click Add or Update to save the custom prompts.
These prompts appear as suggestions when creating a new Ticket with the selected Agent assigned, helping guide Agent interactions efficiently.
Define key-value pairs (mark as mandatory if needed).
Meta Data
Optionally, enter additional key-value configurations.
Set the maximum number of tokens for responses (e.g., 1000).
Knowledge Sources
Optionally, click the plus icon () to connect the Agent to a knowledge source such as a vector database collection. This allows the Agent to retrieve and use information from previously uploaded documents stored in a vector database. Complete the fields:
Vector DB: Select a previously created vector database to connect as a knowledge source.
Collections: Choose one or more document collections within the VectorDB relevant to the Agent (required if Vector DB is selected).
Description: Enter a brief summary of the knowledge source’s purpose or contents (optional).
Meta Data
Add custom key-value metadata to the Agent.
Quick Deploy: Automatically sets up everything needed to run your Agent: it creates a DuploCloud Service, deploys a pod that runs the Agent container, and exposes it through a load balancer listener using the port specified during Agent creation.
Advanced: Allows full control over deployment settings, including network, scaling, and service options.
Proceed through the remaining steps to complete the deployment, following the prompts based on whether you selected Quick Deploy or Advanced. Monitor the deployment status on the Deployments tab.
If it doesn’t exist, click Add. The Add MetaData pane displays.
In the Key field, enter prompt_suggestions.
Update or enter new prompt suggestions in the Value field as a JSON-style array. For example:
Name
Enter a name for the Agent.
Agent Type
Select Prebuilt.
Docker Image
Enter the full image path (e.g., registry/myagent:latest).
Base64 Docker Registry Credential (Optional)
Enter credentials if needed.
Port
Enter the port exposed by the container.
Protocol
Select the network protocol your Agent uses for communication (e.g., http, https, or grpc). This determines how external services connect to the containerized Agent.
Token Limit
Set the token output limit.
Name
Enter a name for the Agent.
Agent Type
Select Dynamic.
Prompt
Enter the initial instruction or context that guides the Agent’s behavior and responses.
Tools
Select one or more registered Tools for the Agent to use. For more about using Tools, see the DuploCloud Tools documentation.
Provider
Select the Large Language Model (LLM) that will power the Agent (e.g., bedrock or Other.
Model
Choose the specific version or configuration of the selected LLM to use for this Agent.
Temperature
Set the randomness of responses (e.g., 0 for deterministic behavior).
Builder Docker Image
This field is prepopulated based on your Agent configurations.
Timeout
Set a timeout for the build job (e.g., 0 for unlimited).
Tools
Select one or more registered tools for the Agent to use. See the Tools documentation for more details.
Build Environment Variables
Define environment variables to initialize tool behavior.
Example:
Key: init_code
Value: from duplo_custom_tool import ExecuteKubectlCommandTool; tool = ExecuteKubectlCommandTool(); print(isinstance(tool, BaseTool)); print(type(tool))
Mandatory
Check if the variable is required.
Custom Build Variables
(Optional) Add any custom key/value pairs for build-time configuration.
Name
Provide a name for this Agent registration.
Instance ID
Enter the ID of the deployed instance (created during deployment).
Allowed Tenants
Select the tenants where this Agent is allowed to operate.
Endpoint
The service endpoint for the deployed Agent (prepopulated).
Path
The endpoint path that handles requests. You can retrieve this from the Agent's registration info, if needed: On the Register tab, click the menu icon () next to the Instance and select Edit. Copy the path from the Path field.
Headers
Optional key/value pairs to pass custom headers during API calls.
kubernetes-agent Agent displayed
Environment Variables
Token Limit
["list all running pods", "show CPU and memory usage for my pods", "display recent events in the cluster"]Meta Data: Add key-value pairs to filter or target specific content in the knowledge source (optional).
