agent-design
AI Agent Memory Architecture Designer
Designs memory systems for AI agents with persistence, retrieval, and context management strategies.
memory-systems agent-architecture context-management
prompt
You are an expert in designing memory architectures for AI agents. Your task is to analyse the given agent requirements and design a comprehensive memory system. ## Agent Requirements [Paste agent description and use case here] ## Design the following memory components: ### 1. Memory Types - **Working Memory**: What information the agent needs during active tasks - **Episodic Memory**: How to store and retrieve past interactions and experiences - **Semantic Memory**: Long-term knowledge storage and organisation - **Procedural Memory**: Task patterns, workflows, and learned behaviours ### 2. Storage Architecture - **Data Structures**: Specific formats for each memory type - **Persistence Layer**: Database schemas, file formats, or vector stores - **Indexing Strategy**: How to efficiently query different memory types - **Memory Hierarchy**: Hot, warm, and cold storage tiers based on access patterns ### 3. Retrieval Mechanisms - **Context Retrieval**: How to pull relevant memories based on current situation - **Similarity Search**: Methods for finding related past experiences - **Temporal Queries**: Accessing memories by time periods or sequences - **Cross-Reference Logic**: Connecting memories across different types ### 4. Memory Management - **Retention Policies**: What to keep, compress, or delete over time - **Conflict Resolution**: Handling contradictory information - **Memory Consolidation**: Merging similar experiences or updating knowledge - **Privacy Controls**: Sensitive information handling and user data boundaries ### 5. Implementation Details - **Technology Stack**: Recommended databases, vector stores, and caching layers - **API Design**: How other agent components interact with memory - **Performance Metrics**: Latency, storage efficiency, and retrieval accuracy targets - **Failure Modes**: What happens when memory systems are unavailable Provide specific technical recommendations, not abstract concepts. Include code snippets or configuration examples where helpful.
Use this when building AI agents that need persistent memory across sessions. Works well with Claude, GPT-4, and Gemini for designing both simple chatbot memory and complex multi-agent systems with shared knowledge stores.