ReAct (Reasoning and Acting) Agent behaviour implementation.

ReAct is a prompting paradigm where AI agents interleave:

• Reasoning: Thinking about what to do next
• Acting: Using tools to gather information or perform actions
• Observing: Processing results to inform the next step

This module implements Nous.Agent.Behaviour with enhanced capabilities:

• Structured planning with facts survey
• Built-in todo list management
• Note-taking for observations
• Loop prevention (warns on duplicate tool calls)
• Mandatory final_answer for task completion

Built-in Tools

• plan - Create structured plan before taking action
• note - Record observations and insights
• add_todo - Track subtasks
• complete_todo - Mark tasks done
• list_todos - View current todos
• final_answer - Complete the task (required)

Example

agent = Agent.new("openai:gpt-4",
  behaviour_module: Nous.Agents.ReActAgent,
  tools: [&search/2, &calculate/2]
)

{:ok, result} = Agent.run(agent,
  "Find the oldest F1 driver and when they won their first championship"
)

Based on Research

This implementation draws from:

• "ReAct: Synergizing Reasoning and Acting in Language Models" (Yao et al., 2023)
• HuggingFace smolagents toolcalling_agent patterns