AI Agent Memory & State Management Architect

You are an expert AI agent memory architect. I need you to design a comprehensive memory and state management system for my AI agent. Context about my agent: - Agent type: [describe your agent - chatbot/coding assistant/research agent/etc.] - Expected conversation length: [short/medium/long sessions] - Key information to remember: [user preferences/task context/facts/decisions] - Infrastructure: [local/cloud, budget constraints] Please design a memory system covering: 1. **Short-term Memory (Working Memory)** - Context window management strategy - What to keep vs. summarize vs. drop - Token budget allocation 2. **Long-term Memory (Persistent)** - Storage backend recommendation (vector DB, KV store, graph DB) - Embedding model selection - Retrieval strategy (semantic search, recency-weighted, hybrid) - Memory consolidation: how to merge/compress old memories 3. **Episodic Memory** - How to store and recall specific past interactions - Session boundary detection - Cross-session continuity 4. **Procedural Memory** - Learned skills and tool-use patterns - Self-improvement loops 5. **Implementation Plan** - Architecture diagram (in text/mermaid) - Key code patterns or pseudocode - Estimated storage/cost for 10K, 100K, 1M interactions Provide concrete, actionable recommendations with specific tool/library choices. Include trade-offs for each decision.