Tencent Cloud Open-Sources Agent Memory Hub for Team-Wide AI Agent Collaboration

Tencent Cloud has released TencentDB Agent Memory v2.0, an open-source memory management system for AI coding agents. The system addresses a key limitation in multi-agent workflows by allowing agents to reuse context and learnings from previous sessions without redundant explanation, while maintaining privacy controls. The platform organizes knowledge into four asset types—Chat Memory, Skill, LLM-Wiki, and Code-Graph—each with versioning and permission management. Memory is hierarchically structured in layers: Chat Memory saves conversations at level L0 and refines them through async pipelines into higher levels (L1 Atom, L2 Scenario, L3 Core/Persona), enabling fast context retrieval while maintaining detail availability. A Memory Proxy supports both OpenAI and Anthropic API protocols, and visibility controls allow sharing between teammates while keeping sensitive data private. The MIT-licensed project runs self-hosted via Docker with support for Linux AMD64 and ARM64 architectures. The v2.0 release adds skill archiving, scheduled repository syncing, system admin tools, bilingual interface support, and a Cost Guard feature for model assignment optimization. Integrations include OpenClaw, Hermes, Claude Code, and CodeBuddy. Self-reported PersonaMem accuracy improved from 48% to 76% in v2.0.