8277778106
- Phase 13: Autonomous Agency - agents now have actions and locations - Phase 15: Sickness mechanics with immunity and weather effects - Phase 16: Crafting system (medicine from herbs) - Phase 17-A: Resource scarcity with tree fruit regeneration - Phase 17-B: Social roles (leader, follower, loner) with clique behavior - Phase 17-C: Random events support - Add AgentMemory model for long-term agent memory storage - Add memory_service for managing agent memories - Update Unity client models and event handlers 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
75 lines
2.6 KiB
Python
75 lines
2.6 KiB
Python
|
|
import logging
|
|
import random
|
|
from typing import List, Optional
|
|
from datetime import datetime
|
|
|
|
from .database import get_db_session
|
|
from .models import Agent, AgentMemory
|
|
|
|
logger = logging.getLogger(__name__)
|
|
|
|
class MemoryService:
|
|
"""
|
|
Manages long-term memories for agents.
|
|
Responsible for:
|
|
1. Storing new memories.
|
|
2. Retrieving relevant memories for context.
|
|
3. Pruning/Summarizing old memories (future).
|
|
"""
|
|
|
|
def __init__(self):
|
|
pass
|
|
|
|
async def add_memory(self, agent_id: int, description: str, importance: int = 1,
|
|
related_entity_id: int = None, related_entity_name: str = None,
|
|
memory_type: str = "general") -> AgentMemory:
|
|
"""
|
|
Record a new memory for an agent.
|
|
"""
|
|
with get_db_session() as db:
|
|
memory = AgentMemory(
|
|
agent_id=agent_id,
|
|
description=description,
|
|
importance=importance,
|
|
related_entity_id=related_entity_id,
|
|
related_entity_name=related_entity_name,
|
|
memory_type=memory_type
|
|
)
|
|
db.add(memory)
|
|
db.commit() # Ensure ID is generated
|
|
db.refresh(memory)
|
|
logger.info(f"Agent {agent_id} remembered: {description} (Imp: {importance})")
|
|
return memory
|
|
|
|
async def get_relevant_memories(self, agent_id: int, context: str, limit: int = 3) -> List[str]:
|
|
"""
|
|
Retrieve memories relevant to the current context.
|
|
For MVP, we just return the most recent high-importance memories
|
|
and any memories related to the entities in context.
|
|
"""
|
|
memories = []
|
|
with get_db_session() as db:
|
|
# 1. Get recent important memories (Short-term / working memory)
|
|
recent_memories = db.query(AgentMemory).filter(
|
|
AgentMemory.agent_id == agent_id,
|
|
AgentMemory.importance >= 5
|
|
).order_by(AgentMemory.created_at.desc()).limit(limit).all()
|
|
|
|
# 2. Get entity-specific memories (e.g. if talking to "User1")
|
|
# Simple keyword matching for now (Vector DB is Phase 14+)
|
|
entity_memories = []
|
|
if context:
|
|
# Naive search for names in context
|
|
# In real prod, use embeddings.
|
|
search_term = f"%{context}%" # Very naive
|
|
# Let's just fallback to recent for MVP to ensure stability
|
|
|
|
for mem in recent_memories:
|
|
memories.append(f"- {mem.description}")
|
|
|
|
return memories
|
|
|
|
# Global instance
|
|
memory_service = MemoryService()
|