> ## Documentation Index > Fetch the complete documentation index at: https://praison.ai/docs/llms.txt > Use this file to discover all available pages before exploring further. # Memory • AI Agent SDK > Memory: A single-file memory manager covering: # Memory > Defined in the [**memory**](../modules/memory) module. AI Agent A single-file memory manager covering: * Short-term memory (STM) for ephemeral context * Long-term memory (LTM) for persistent knowledge * Entity memory (structured data about named entities) * User memory (preferences/history for each user) * Quality score logic for deciding which data to store in LTM * Context building from multiple memory sources * Graph memory support for complex relationship storage (via Mem0) Config example: \{ "provider": "rag" or "mem0" or "mongodb" or "none", "use\_embedding": True, "short\_db": "short\_term.db", "long\_db": "long\_term.db", "rag\_db\_path": "rag\_db", # optional path for local embedding store "config": \{ "api\_key": "...", # if mem0 usage "org\_id": "...", "project\_id": "...", # MongoDB configuration (if provider is "mongodb") "connection\_string": "mongodb://localhost:27017/" or "mongodb+srv://user:[pass@cluster.mongodb.net](mailto:pass@cluster.mongodb.net)/", "database": "praisonai", "use\_vector\_search": True, # Enable Atlas Vector Search "max\_pool\_size": 50, "min\_pool\_size": 10, "max\_idle\_time": 30000, "server\_selection\_timeout": 5000, # Graph memory configuration (optional) "graph\_store": \{ "provider": "neo4j" or "memgraph", "config": \{ "url": "neo4j+s\://xxx" or "bolt://localhost:7687", "username": "neo4j" or "memgraph", "password": "xxx" } }, # Optional additional configurations for graph memory "vector\_store": \{ "provider": "qdrant", "config": \{"host": "localhost", "port": 6333} }, "llm": \{ "provider": "openai", "config": \{"model": "gpt-4o-mini", "api\_key": "..."} }, "embedder": \{ "provider": "openai", "config": \{"model": "text-embedding-3-small", "api\_key": "..."} } } } Note: Graph memory requires "mem0ai\[graph]" installation and works alongside vector-based memory for enhanced relationship-aware retrieval. ## Constructor No description available. No description available. ## Methods Combine multiple sub-metrics into one final score, as an example. Store in short-term memory with optional quality metrics Search short-term memory with optional quality filter Completely clears short-term memory. Store in long-term memory with optional quality metrics Search long-term memory with optional quality filter Clear local LTM DB, plus Chroma, MongoDB, or mem0 if in use. Delete a specific short-term memory by ID. Delete a specific long-term memory by ID. Delete a specific memory by ID. Delete multiple memories by their IDs. Delete memories matching a search query. Save entity info in LTM (or mem0/rag). Filter to items that have metadata 'category=entity'. If you only want to drop entity items from LTM, you'd do a custom If mem0 is used, do user-based addition. Otherwise store in LTM with user in metadata. If mem0 is used, pass user\_id in. Otherwise fallback to local filter on user in metadata. Generic search method that delegates to appropriate specific search methods. Clear all user-based info. For simplicity, we do a full LTM reset. Store task output in memory with appropriate metadata Merges relevant short-term, long-term, entity, user memories Fully wipes short-term, long-term, and any memory in mem0 or rag. Calculate quality metrics using LLM Store quality metrics in memory Search with quality filter Get all memories from both short-term and long-term storage Get the LearnManager for continuous learning capabilities. Get learning context suitable for injection into system prompt. ## Source `praisonaiagents/memory/memory.py` at line 126 *** ## Related Documentation