AgentDB — персистентная память для AI-агентов

ai-tooling

agentdb-memory-patterns is a Claude Code skill that implements persistent memory patterns for AI agents using AgentDB and its ReasoningBank integration. It covers four memory tiers — session history, long-term fact storage, semantic vector search with Maximal Marginal Relevance, and hierarchical memory with importance-based consolidation. The skill includes CLI workflows via npx agentdb@latest for database initialization, MCP server setup inside Claude Code, and learning plugin generation from templates such as decision-transformer, q-learning, and actor-critic; benchmarks show pattern search up to 12,500× faster than traditional approaches. Designed for developers building stateful agents, conversational systems, or intelligent assistants that must retain context across sessions — requires AgentDB v1.0.7+ and Node.js 18+, available standalone or via agentic-flow.