5.3.1 Create an Agent Profile

Create an agent profile to define a reusable AI agent in MicroTx Workflows. An agent profile provides a centralized place to configure an agent's role, instructions, prompt variables, tools, MCP servers, LLM profile, model parameters, memory behavior, execution limits, capabilities, and guardrails.

Use agent profiles to run agents within workflows through Agentic Tasks or as conversational agents through the chat API. You can port templates defined for external agents and execute it within MicroTx Workflows.
Before creating an agent profile, ensure that the required LLM profile, tool configurations, and MCP server definitions are already available.
  1. Open the navigation menu and click Agentic AI.
  2. Click the Agent Profile tab.

    The Agent Profile Definitions list page opens. All the agent profiles that you have defined are displayed in a table.

  3. Click add a new building block.
    The New Agent Profile Definition dialog box appears.
  4. Enter the following information.
    • Name: Enter a unique and descriptive name. This name is used to reference the agent profile definition in an Agentic Task in a workflow or through the chat conversation API using the agentName parameter. The name can be up to 128-characters long. Use only letters, numbers, underscores (_), and hyphens (-). Spaces and other special characters are not supported.
    • Description: Optional. Provide a brief description of this agent profile.
    • Role: Specify the role of the agent to be included in the execution prompt. This field helps establish the agent’s high-level behavior, responsibilities, boundaries, and tone as shown in the following example. Keep the content concise and stable so it consistently guides the agent during execution. The maximum supported length is 200000 characters.
      You are an expert Oracle Database Assistant. Your primary function is to help users interact with the database by translating their natural  language requests into valid SQL queries. You are precise, knowledgeable,and cautious, especially with data modification commands.
    • Instruction: Enter detailed instructions to define the agent's behavior during execution. Describe what the agent should do when invoked, how it should respond, how it should use available tools, and any behavioral rules or constraints it must follow while interacting with workflow inputs, tools, or users. Keep the instructions clear, specific, and action-oriented so the agent behaves consistently as shown in the following example. The maximum supported length is 400000 characters.
      Your  primary task is to parse the user's request and convert it into a valid  Oracle SQL query.
      1. Analyze the user's request to understand their  goal (e.g., retrieve data, describe a table, count rows).
      2. Construct the appropriate SQL query using Oracle syntax.
      3. Present the results  from the database to the user in a clear and easy-to-understand format. For example, a markdown table for data.
    • Capabilities: Select one or more of the following options to define the agent's operation modes.
      • Workflow: Allows agent to run within MicroTx Workflows, such as through an Agentic Task, while a workflow manages the execution.
      • Conversational: Allows agent to run as a conversational agent in free-form chat interactions through the conversation API.
    • LLM Profile: Select an LLM profile that powers the agent's reasoning and language tasks.

      Note: Agentic task does not support Cohere models.

    • Use Memory: Optional. This setting applies only to conversational capability and it does not change behavior for workflow or agentic task execution. It controls whether a conversational agent uses prior chat context, including recent messages and summary, to answer new prompts in the same chat session. When enabled, the agent keeps contextual continuity across turns, so responses can reference earlier parts of the conversation. When disabled, the agent responds using only the current input and profile instructions; prior session context is not used to generate response. It is enabled by default.

      Caution:

      Disabling memory may reduce response relevance in multi-turn conversations since earlier context is not considered.
    • Max Messages: Optional. Sets the maximum number of chat messages kept in the agent's in-memory context window during execution. When the limit is exceeded, older messages are evicted and only the most recent messages are sent to the model. This helps control token usage and cost. The default value is 20.
    • Prompt Variables: If the prompt contains any variables, add the variables and specify the values that replace these variables at runtime.
    • Guardrails - Stop Words: Define runtime stop-word rules that filter both requests and responses. For each stop-word rule, specify:
      • words: a case-insensitive list of terms or phrases to detect or match.
      • scope: specify where the system applies the rule
        • INPUT: scan text before sending it to the model.
        • OUTPUT: scan the model's response before returning it.
        • BOTH: scan both input and output.
      • action: specify what the system does when it finds a match:
        • MASK: replace the matched text with ***.
        • FAIL: stop execution and raise a guardrail violation error.

      At runtime, the system evaluates all rules. It runs input guardrails before the LLM call and output guardrails after the LLM generates a response. It matches content using case-insensitive string matching. If you omit or provide an invalid scope or action, the system does not enforce that rule. If you omit words or set it to an empty list, the rule has nothing to match.

    • Temperature: Optional. Controls randomness of the generated response. Lower values generate more deterministic responses. Higher values generate more varied responses. The default value is 0.2.
    • Max Tokens: Optional. Sets the maximum number of tokens that the model can generate. The default value is 1024. Optional. Sets the maximum number of tokens the model can generate in a response. The default value is 1024.
    • Top K: Optional. Number of highest-probability tokens considered during generation. The default value is 40.
    • Top P: Optional. Controls nucleus sampling value. The default value is 0.9.
    • Max Tool Calls: Optional. Limits the number of tool invocations per session. This prevents runaway invocation loops and constrains execution cost and side effects. The default value is 10.
    • MCP Servers: Optional. Select one or more Model Context Protocol (MCP) servers that the agent can access and use for executing tasks or accessing resources.
    • Tools: Optional. Select one or more tool configurations that you have created previously in MicroTx Workflows. The agent accesses and invokes the specified tools during workflow execution.
  5. Click Submit.
Your new definition appears in the list of available agent profile definitions. You can invoke the agent profile by name in workflows or tasks.

Example

The following JSON code provides sample values for parameters of an agent profile.

{
  "name": "orderProcessorAgent",
  "description": "Handles order-related workflow and chat queries",
  "role": "Process and respond to all order fulfillment workflow tasks",
  "instruction": "Always validate input, check inventory before approving order, escalate exceptions",
  "llmProfile": {
    "model": "gpt-4",
    "vendor": "OpenAI"
  },
  "tools": [
    "OrderDBLookupTool",
    "EmailSender"
  ],
  "mcpServers": [
    "inventoryMCP",
    "shippingMCP"
  ],
  "memory": true,
  "maxMessages": 30,
  "maxToolCalls": 5,
  "capabilities": [
    "WORKFLOW",
    "CONVERSATIONAL"
  ],
  "promptVariables": {
    "companyBrand": "MicroTx",
    "region": "APAC"
  },
  "temperature": 0.5,
  "maxTokens": 2048,
  "topK": 50,
  "topP": 0.9,
  "maxCompletionTokens": 512,
  "guardrails": {
    "stopWords": [
      {
        "words": [
          "password"
        ],
        "scope": "INPUT",
        "action": "MASK"
      }
    ]
  }
}