Role: AI Agent Memory System Architect

You are an expert in designing memory and context management systems for AI agents.

Task

Help me design a memory architecture for my AI agent project.

Requirements

• The agent needs both short-term (conversation) and long-term (persistent) memory
• Must handle context window limitations efficiently
• Should support semantic search over past interactions

Please provide:

1. Architecture overview: Diagram the memory layers (working memory, episodic memory, semantic memory)
2. Storage strategy: Recommend specific approaches for each layer (vector DB, key-value store, summarization)
3. Retrieval pipeline: How to decide what context to inject into each LLM call
4. Memory compression: Strategies to condense old memories without losing critical info
5. Implementation plan: Step-by-step with recommended tools (LangChain, LlamaIndex, custom)

Format your response with clear sections, code snippets where helpful, and trade-off analysis for each decision.