Agent Memory and State Management System

# Agent Memory and State Management System

You are an AI systems architect. Design a complete memory and state management system for [AGENT_SYSTEM_NAME].

## System Context

- Agent system: [AGENT_SYSTEM_NAME]
- Number of agents: [AGENT_COUNT]
- Session length: [SESSION_DURATION]
- Cross-session persistence required: [YES_OR_NO]
- Multi-user or single-user: [USER_TYPE]

## Memory Taxonomy

**Working Memory** (current session only)
- What it stores: active task context, intermediate results, current user intent
- Capacity limit: [WORKING_MEMORY_LIMIT]
- Eviction policy: [EVICTION_STRATEGY]

**Episodic Memory** (past interactions)
- What it stores: conversation history, decisions made, outcomes observed
- Retention period: [RETENTION_PERIOD]
- Structure: [EPISODIC_SCHEMA]

**Semantic Memory** (learned knowledge)
- What it stores: facts, user preferences, domain knowledge
- Update trigger: [UPDATE_CONDITIONS]
- Confidence scoring: how certainty is tracked per fact

**Procedural Memory** (how-to knowledge)
- What it stores: successful workflows, proven prompt patterns
- When to apply: [RETRIEVAL_CONDITIONS]

## Retrieval Architecture

Define how agents query memory:
- Similarity search for relevant past episodes
- Keyword retrieval for specific facts
- Recency weighting formula
- Confidence threshold for using vs. ignoring a memory

## Conflict Resolution

When two agents write conflicting information to shared memory:
1. Detection: how the conflict is identified
2. Resolution strategy: [RESOLUTION_RULES]
3. Audit: how conflicts are logged for review

## Persistence Layer

Specify the storage backend, serialization format, and backup strategy for production deployment.