Tencent Open-Sources Agent Memory as a Governed Team Asset

TencentDB-Agent-Memory added roughly 1,090 stars in a day on a premise that differs from mem0 and Zep: memory as a team asset rather than a per-agent store. Conversations refine through four layers, from raw transcript to atoms of fact, project scenarios and long-term personas, and surface as four asset types — chat memory, skills, wiki and a code graph. Assets bind to specific agents through an ACL instead of being injected wholesale. The framing is the contribution; a memory hub concentrates everything your agents have seen, so the access layer is load-bearing security. Self-described active beta.