Configure the Agentic AI Thinking Pattern

AI agents can use different thinking patterns for reasoning. Before you can create an AI agent, you need to create the thinking pattern the agent will use.

  1. In the left navigation pane, select Projects.
  2. Select the Simple Expense Approval project that you created.
  3. In the left navigation pane, select AI Agents AI Agent icon.
  4. In the Agent patterns box, click Add.

    The Create pattern panel is displayed.

  5. In Create pattern, configure required fields for your pattern:

    Create pattern screen with fields for name ("ReAct for Simple Expense Approval Agent"), identifier ("REACTPATTERN_FOR_AGENTS"), Version ("01.00.000"), pattern recipe ("ReAct" selected), and an empty description box. At the bottom are Cancel and Add buttons.

    • Name: Name for the thinking pattern. For our tutorial, enter the name ReAct for Simple Expense Approval Agent. Currently, there are two patterns available in Oracle Integration. ReAct and Plan and Execute. You might want to identify your pattern with those names for easier understanding.
    • Identifier: Specify REACTPATTERN_FOR_AGENTS.
    • Version: Accept the default value of 01.00.0000 or change it, if needed. You can create different versions of the same pattern. For example, you create the Simple Decision pattern with two different versions:
      • Version 01.00.0000 uses the ReAct pattern recipe.
      • Version 02.00.0000 uses the Plan pattern recipe.
    • Pattern recipe: Value sent to the LLM as a system prompt to define how the agent reasons. Select the ReAct pattern for the tutorial. A pattern recipe is implemented as an integration that contains the thinking pattern for the AI agent. Oracle Integration ships with two ready-to-use patterns (ReAct and Plan) and two OCI Generative AI patterns:
      • ReAct for OCI GenAI: Reasoning and action pattern for OCI Generative AI models. See Generative AI.
      • ReAct: Reasoning and action pattern for non-OCI Generative AI models. The AI agent alternates between reasoning and acting. The AI agent updates its plans based on new information that it learns. The AI agent prompts the LLM for the next action, acts, learns, and then repeats.
      • Plan GenAI: Planner for plan and react pattern for OCI Generative AI models. See Generative AI.
      • Plan: Planner for plan and react pattern for non-OCI Generative AI models. The AI agent prompts the LLM to generate a complete plan, then executes that plan step by step.

      ReAct and Plan are two separate patterns that must be orchestrated in sequence, respectively.

  6. Click Create.
    The Pattern details page is displayed.
  7. Configure the Pattern and LLM.
    1. Leave the following fields with the default number:
      • Max iterations: Leave Max iterations as the default number. This limits the number of reasoning and action steps the agent can take before providing a final answer or exiting. This field prevents infinite loops.
    2. In the Guidelines box, enter the following guidelines for the ReAct thinking pattern:

      Screen for the ReAct for Simple Expense Approval Agent pattern, showing max iterations set to 10 and temperature set to 0. The guidelines box explains how to follow the ReAct agent pattern with steps for think, action, observe, and repeat. On the right, LLM Connections displays "LLM_Connection."

      Strictly follow all guidelines. It is mandatory that you reflect on all guidelines provided.

You should follow the ReAct agent pattern (Reason + Act) when generating a response.
Pattern:
Think - using your internal reasoning.
Action - Invoke tools to get information, including additional human input.
Observe - Process tool response using your internal reasoning.
Repeat - Repeat until finished.

At each step you provide your think, action, and observe rationale.

You should always attempt to get the most relevant information based on tool use.

Do not guess or infer tool arguments except for the following reasons:

You are given specific guidance by the tool on inferring a tool argument.

You have previous context to make a reasonable assumption.
    3. Click Save.
    4. Under LLM Connections, click Edit icon to configure the LLM connection.
      A new LLM connection is created and is displayed.
    5. Configure connection information for your LLM.

      LLM_Connection configuration page. The left displays properties for the connection, including Base URL "https://api.openapi.com ," model "gpt-4o-mini," and a masked API key for API Key Based Authentication. Configuration progress shows 100%. On the right, a "Save changes?" panel indicates there are no active integrations, but the inactive integration "ReAct Pattern Intg (1.0)" is configured. "Close" and "Save" buttons appear at the bottom.

      • Base URL: URL to connect to your LLM. For example: https://api.openai.com.
      • Model: Model to use for your AI agent. For example: gpt-4o-mini
      • API Key Based Authentication: Specify your API key to connect to the LLM.
    6. Click Test to test that your connection works, then click Save.

      When you save, Oracle Integration creates an integration that implements the ReAct pattern called ReAct Pattern Intg in the Integrations section. This system-created integration is automatically activated for the agent.

  8. Click Back Back icon to return to the Pattern details page.
  9. Click Back Back icon to return to the AI Agents page.
  10. Activate the thinking pattern. The pattern must have the status Active before you can specify it in your agent.
    1. In the Agent patterns section, next to ReAct for Simple Expense Approval agent, click Actions Actions icon, and select Activate.

After your thinking pattern is created for your AI agent, Oracle Integration automatically creates the ReAct Pattern Intg integration in your project. Select the Integration section to see it. The integration is system-generated and is required for your AI agent. You don't need to change or edit it.

The LLM connection is also automatically created for you.

Next step: Add and Configure the AI Agent