TencentDB-Agent-Memory: Cut the Cost of Agents Relearning from Scratch Every Time with Structured Assets

This tool focuses on reducing the cost of agents relearning projects from scratch every time. Beyond simply storing existing conversation logs, it transforms repetitive tasks and context into structured assets for reuse. It is designed so that new sessions or other agents do not need to re-verify previous decisions and documents.

The core features consist of four components: Chat Memory extracted from conversations, reusable Skills, document-based Wiki, and CodeGraph which analyzes code structure. Chat Memory refines user preferences and decision histories by hierarchy, while Skills are executable experiences equipped with version control and execution rules. Wiki and CodeGraph organize the relationships between documents and code in graph form to help agents find only the necessary information accurately.

Unlike existing RAG which answers 'what can be found,' this system controls 'who can use which version of an asset.' Through a portable asset structure independent of agent frameworks, various agents such as Claude Code, Codex, and CodeBuddy can share the same memory server. A key feature is zero-code integration, which works immediately by simply changing the Base URL without installing plugins or MCP servers.

The Memory Hub panel allows management of asset ownership, versions, and access permissions at the team level. Personal assets are private by default and can be shared with the entire team or specific agents after review. It also provides cold-start support to import existing codebases, documents, and past conversation sessions and convert them into assets immediately.