[WIP] Refactor agent classes to data classes and organize modules

[Copilot]

Refactoring Plan: Agent Classes to Data Classes with Ray Parallelization

Phase 1: Convert Agent Classes to Data Classes

• Create data class for base agent with all attributes
• Create data class for page agent extending base agent data
• Create data class for fake agents
• Remove all methods from agent classes (keep only data)

Phase 2: Extract Methods to Functional Modules

• Create agent_actions.py module for CPU-bound operations (post, comment, share, reaction, follow)
• Create agent_llm_operations.py module for GPU-bound LLM operations
• Create agent_data_operations.py module for data retrieval operations
• Create agent_utils.py module for utility functions
• Organize functions into logical groups

Phase 3: Ray Integration

• Add ray to requirements
• Create ray remote functions for CPU-bound operations
• Create ray remote functions for GPU-bound LLM operations
• Update client to use ray.get() for parallel execution
• Configure ray with appropriate resources (CPU/GPU)

Phase 4: Testing and Validation

• Test agent creation and initialization
• Test individual agent actions
• Test simulation execution
• Validate parallelization improvements

Notes:

• GPU-bound functions: post, comment, share, reaction (with LLM calls), select_action (with LLM), emotion_annotation, news (for page agents)
• CPU-bound functions: follow, read, search, data retrieval/API calls
• Keep minimal changes to maintain backwards compatibility

Original prompt

Rewrite the agent classes to become data classes. Make the method functions in one or more neatly organized modules.

Implement parallelization with ray: use agent data classes and make the functions use them efficiently. Differentiate from CPU bound and GPU bound functions (the ones interacting with LLMs).

---

💡 You can make Copilot smarter by setting up custom instructions, customizing its development environment and configuring Model Context Protocol (MCP) servers. Learn more Copilot coding agent tips in the docs.