Elite Longterm Memory

# Elite Longterm Memory Multi-layered AI agent memory system combining session state, vector search, git-notes, curated archives, and optional cloud backup — so your agent never loses context across sessions. Elite Longterm Memory provides a comprehensive 6-layer memory architecture for AI coding agents. It combines hot session state (Write-Ahead Log), warm vector search (LanceDB), cold structured knowledge (git-notes), curated human-readable archives (MEMORY.md), optional cloud backup (SuperMemory), and automatic fact extraction (Mem0) into a single unified system. Your agent retains context, decisions, preferences, and lessons learned across sessions and compaction events. ## How It Works - **Hot RAM**: (SESSION-STATE.md) stores active working memory using a Write-Ahead Log protocol — state is written before the agent responds, ensuring nothing is lost on crashes or compaction. - **Warm Store**: (LanceDB vectors) enables semantic search across all stored memories with auto-recall. - **Cold Store**: (git-notes) preserves structured decisions and learnings in a branch-aware knowledge graph. - **Curated Archive**: (MEMORY.md + daily logs) provides human-readable long-term storage. Optional layers include SuperMemory for cloud backup and Mem0 for automatic fact extraction with 80% token reduction. ## Key Features - **Write-Ahead Log Protocol**: State is persisted before the agent responds, guaranteeing durability even through crashes and context compaction - **Semantic Vector Search**: LanceDB-powered auto-recall injects relevant prior context into conversations automatically - **Git-Notes Knowledge Graph**: Branch-aware structured storage for decisions, learnings, and architectural context - **Daily Logs + MEMORY.md**: Human-readable curated archives with daily activity logs and distilled long-term wisdom - **Cloud Backup (Optional)**: SuperMemory API enables cross-device sync and chat with your knowledge base - **Auto Fact Extraction (Optional)**: Mem0 automatically extracts preferences, decisions, and facts from conversations with 80% token reduction - **Memory Hygiene Workflow**: Built-in weekly maintenance protocol to audit, clean, and consolidate memory ## Requirements - **LLM API Key (OpenAI)**: Powers semantic vector search in LanceDB. Required for the warm store layer - **SuperMemory API Key (Optional)**: For cloud backup and cross-device sync - **Mem0 API Key (Optional)**: For automatic fact extraction from conversations ## Use Cases - **Long-Running Projects**: Maintain context across dozens of coding sessions without re-explaining decisions - **Team Knowledge**: Build a persistent knowledge base of project decisions, patterns, and lessons learned - **Preference Retention**: Your agent remembers coding style preferences, framework choices, and workflow patterns - **Error Prevention**: Log mistakes and lessons so the agent never repeats the same errors - **Cross-Session Context**: Seamlessly resume work where you left off, even after agent restarts or context compaction ## Installation Install via: `npx clawhub@latest install elite-longterm-memory`

Установка

Похожие навыки