simplemem
par nantes
Search & ResearchRésumé
Efficient Lifelong Memory for LLM Agents - semantic compression, cross-session memory, and intent-aware retrieval.
Description
Ce que fait ce skill
SimpleMem gives LLM agents persistent memory across sessions by compressing interactions into compact units and retrieving them based on intent. It uses a multi-index approach combining semantic, lexical, and symbolic layers. Without an OpenAI key, it falls back to JSON-based keyword search.
Unlike naive context stuffing, SimpleMem compresses and indexes memories so retrieval stays fast and relevant even as history grows.
Quand l'utiliser
- Remembering user preferences across chat sessions
- Storing project context for long-running agent tasks
- Reducing repeated context in multi-turn conversations
- Retrieving relevant past interactions during planning
- Compressing conversation history to stay within token limits
Exemple de workflow
User asks: Remember that I prefer dark mode and always use TypeScript
- 1
Calls simplemem add with the preference string and a user ID
- 2
SimpleMem compresses and stores the memory unit in persistent storage
- 3
In a later session, agent calls simplemem retrieve with query about user preferences
- 4
SimpleMem runs intent-aware search across semantic and lexical indexes
- 5
Returns ranked memory results with the stored preference
Agent recalls dark mode and TypeScript preference without re-prompting the user
Prérequis
- OPENAI_API_KEY environment variable (optional, required for semantic embeddings)
Installation
clawhub install nantes/simplememExécute la commande CLI, ou colle l'instruction dans Claude Code, Cursor, Codex ou l'agent de code que tu utilises.