Tencent Cloud launches TencentDB Agent Memory v2.0

Tencent Cloud has released the stable 2.0.0 version of TencentDB Agent Memory, an MIT-licensed, self-hosted memory hub designed to help AI coding agents retain and share project context. Launched on August 3, 2026, the system eliminates the need to repeatedly explain project details in new sessions. It operates via three Docker images supporting linux/amd64 and linux/arm64 architectures, targeting solo developers, small teams, and mid-sized organizations looking for shared infrastructure.

The system structures data into four distinct memory assets: Chat Memory, Skill, Wiki, and CodeGraph. Chat Memory processes raw conversations at level L0 and refines them asynchronously into L1 Atom, L2 Scenario, and L3 Core/Persona tiers. Retrieval combines BM25, vector search, and Reciprocal Rank Fusion (RRF) while respecting strict token and timeout budgets. Skills capture reusable procedures with execution steps and validation rules, while Wiki builds structured knowledge pages and CodeGraph indexes code symbols and call relationships.

A key differentiator for developers is the built-in access control list (ACL) governance layer. Memory assets default to private, but can be shared at team or restricted levels, ensuring teammate agents can collaborate without leaking sensitive data. The system includes a Memory Proxy that supports Anthropic and OpenAI protocols, integrating with tools like Claude Code, OpenClaw, Hermes, and CodeBuddy. It runs on default ports 8420, 8125, 8424, and 8096, and offers official SDKs for Python and TypeScript.

The v2.0.0 update introduces features like scheduled CodeGraph repository syncing, forced archiving for Skills, system-admin asset management, English and Chinese panel switching, and a Cost Guard to assign cheaper models to specific agents. Tencent Cloud reports that the system boosted PersonaMem accuracy from 48% to 76%, though this 59% relative improvement remains self-reported. For practitioners, this setup provides a secure, local way to onboard agents to codebases and run impact analyses without relying on third-party APIs.