Step 4: Configuring Subagent and A2A Parameters

Configuring Subagent Parameters

When a Subagent is added to the canvas, a dedicated parameter configuration panel appears. This panel allows you to define the subagent’s behavior, scope, and interactions within the workflow. It enables precise control over how input is handled, what the subagent is expected to do.

For example: You are an Invoice Processing Agent. Extract invoice number, vendor, amount, and date from incoming documents and flag discrepancies.

You can also specify how the input is passed to the agent, and how the agent’s response or extracted data is used in the next steps of the workflow.

The configuration interface is organised into three key tabs:

This structured approach to subagent configuration allows users to build precise, purpose-driven AI agents with role-specific instructions and tool integrations - without writing any code.

Mandatory Parameters

Use this tab to capture the fundamental attributes and prompt settings required to initialize the subagent.

Option Description
*Indicates mandatory field
Subagent Name * Enter a specific name for this subagent. If it is the primary agent, it will default to "New Subagent" but should be renamed to something meaningful that reflects its role within the workflow.

Prompt

This section is critical for guiding the LLM's behavior for this specific subagent.
Role*

Specify the subagent’s functional role or identity the AI agent should adopt for its specific tasks within the workflow. This guides how the agent interprets instructions.

Example: Customer Support Specialist, Financial Analyst.
Scope*

Define the subagent’s area of responsibility or task boundaries within the larger workflow.

Example: Analyze financial statements for Q3 2025.
LLM/SLM

This section allows you to configure the Large Language Model (LLM) or Small Language Model (SLM) that your agent will use for generating or processing responses.

You can select an existing provider and profile, or create a new profile to define credentials and model configurations.

  • Use agent's LLM/SML: Select this to use the global or default LLM/SML settings configured for the parent agent or application.
  • Set a different LLM/SML: Select this to override the default settings and manually specify a different model provider and profile. This option activates the Provider and Profile fields.
Note: For more details, refer Configuring LLM/SLM Provider and Profile Settings, in Configuring Parameters page.
Execution Limits Section, for details refer agenticaibuilder_creatingagents_step2.html#agenticaibuilder_creatingagents_step2__executionlimits

Instructions

Use this tab to provide clear, detailed step-by-step instructions and rules that guide the AI agent's reasoning and execution. These instructions are specific to the subagent's role and scope.

  • Click the Add Instruction + button to define one or more step-by-step instructions.

  • Each instruction should be concise and directive to ensure the AI performs accurately and consistently.

Actions

In this tab you can define the external tools or functions that the subagent is allowed to invoke during execution. This enables the agent to interact with third-party systems or trigger downstream actions

Option Description
v2.1.5+Tools: Add Tool
Click Add Tool, then choose the required tool from the Tool Category dropdown list.
Note: You can find detailed information about each Tool Category and its specific configuration options in the Tools Library page.

Configuring A2A gateway configurations

The A2A Node acts as an A2A Client within your workflow, enabling the agent to delegate tasks and consume specialized capabilities from another remote AI Agent. By leveraging the open Agent2Agent (A2A) Protocol, this node ensures secure, standardized communication and seamless collaboration between AI agents, regardless of their underlying platform or vendor. This capability is essential for creating robust, modular multi-agent systems.

When a A2A subagent is added to the canvas, a dedicated parameter configuration panel appears. This panel allows you to define the A2A subagent’s behavior, scope, and interactions within the workflow. It enables precise control over how input is handled, what the subagent is expected to do.

v2.1.5+A2A Server

The A2A Server in Agentic AI Builder functions as the internal communication hub that enables agents to securely connect, discover, and interact with other agents hosted within the platform ecosystem.

Unlike static APIs, the A2A integration is dynamic and multi-tenant: any newly deployed agent is instantly discoverable and reachable via the A2A protocol, without requiring manual configuration or system restarts.

For more information, refer How A2A Communication Works.

Option Description
*Indicates mandatory field
Name * A user-friendly unique ID assigned to the delegated agent. Example: ItineraryPlannerAgent
Base URL*

The base URL of the remote service. Your sub-agent uses this endpoint to automatically discover the capabilities listed on the remote agent's Agent Card.

For example: https://api.travel-ai.com
Credential Type*

The type of security credential specified by the remote agent. This must be selected before supplying the credential.

Example: Bearer Token Crdential
Credential* The actual authentication value (for example, API key string, tok) required to access the remote agent. The format depends on the scheme defined in the Agent Card.
Note: Defining the Credential is mandatory for secure Agent-to-Agent communication, ensuring only authenticated, trusted interactions occur within enterprise environments.

Saving and Deploying the Agent

After entering all the required parameters, complete the following steps to save and deploy the agent.

Save the agent

  1. Select the folder where the agent configuration will be stored.

    The Folder field displays the selected path (for example, /).

  2. Click Save in the top‑right corner of the builder interface to preserve your configuration.

  3. After saving, the agent is set to the In Progress state by default.

  4. Update the agent status to Ready.

    Only agents marked as Ready are visible in the Agent Catalog and are eligible for deployment.

Deploy the agent

  1. Navigate to the AI Agents page.

  2. Locate the agent, and click the Deploy icon to deploy the agent.

    Note: Only agents in the Ready state can be deployed.

After deployment, the agent becomes available for use in workflows.

How A2A Communication Works

The Sub-Agent communicates with a remote A2A Exposed AI Agent through a structured sequence involving discovery, authentication, and task execution. The process is as follows:

1. Discovery Using the Base URL

The Sub-Agent begins by accessing the remote agent’s Base URL to retrieve its Agent Card.

The Agent Card is a JSON manifest that provides essential metadata, including:

  • The agent’s name and description

  • Its available skills and capabilities

  • The service endpoint for task submission

  • The authentication method required (which determines the appropriate Credential configuration)

This discovery step enables the Sub-Agent to understand how to interact with the remote agent and what operations it can request.

2. Authentication Using Credentials

Based on the authentication method declared in the Agent Card, the Sub-Agent uses the configured Credential to securely authorize communication.

This ensures that only trusted, authenticated agents can send delegated tasks to the remote agent.

3. Task Communication

After successful authentication and capability discovery, the Sub-Agent sends a Task request—the core work unit in A2A communication.

Tasks are submitted via JSON-RPC over HTTP(S) to the remote agent’s service endpoint (typically derived from the Base URL).

This establishes a structured, secure communication channel for delegating and executing work.

Note: You can continue to create A2A sub‑agents that connect to external agents exposed by another A2A server. This functionality remains unchanged from previous versions.

v2.1.5+A2A Server Lifecycle

The A2A Server manages the end-to-end communication lifecycle between agents inside the Agentic AI Builder platform. It ensures secure discovery, tenant isolation, and standardized task execution across a multi-tenant environment.

Note: The A2A Server allows agents created in the Agentic AI Builder to be exposed via the A2A protocol. In the current release, this capability is available only for internal communication between platform services. It is not yet exposed to the public internet or external third‑party networks.

Lifecycle Phases

  1. Discovery Phase: Host and Expose

    • Dynamic Agent Cards: Real-time profiles generated on demand, exposing agent name, description, capabilities (for example, streaming), and endpoints.

    • On-Demand Routing: Communication routes are provisioned only when an agent is active, minimizing overhead and ensuring security.

  2. Authentication and Isolation Phase: Guarding and Verification

    • Tenant Resolution: Requests are mapped to tenants via headers or authorization tokens; safe defaults apply if no tenant is resolved.

    • Data Isolation: Each tenant runs in a dedicated workspace. Even though multiple tenants share the A2A Server, their environments remain fully segregated.

  3. Task Communication Phase: Execution and Response

    • Protocol Translation: Validates payloads and translates them into agent-executable instructions.

    • Flexible Interaction Modes:

      • Sync: Full response returned after task completion.

      • Streaming: Real-time updates during execution (for example, “working”, “input required”, “completed”).

    • Standardized Responses: Results or structured error messages are packaged into a consistent format for the calling service.
Note:

  • Credential Security: Tokens and keys must be stored securely.

  • Multi-Tenant Safety: Isolation guarantees no cross-tenant data leakage.

  • Error Handling: Discovery/authentication failures should trigger safe fallback mechanisms.

  • Scalability: Dynamic routing ensures efficient resource usage as agent count grows.