Tencent Cloud Open-Sources TencentDB Agent Memory v2.0 Team-Level Hub
Tencent Cloud has open-sourced TencentDB Agent Memory v2.0, a self-hosted memory management system designed for AI coding agents with multi-architecture support.
Stock photo for illustration only, not from the actual event
- TencentDB Agent Memory v2.0 is an MIT-licensed, self-hosted open-source system.
- Converts work into four asset types: Chat Memory, Skills, Knowledge, and CodeGraph.
- Reports PersonaMem accuracy increasing from 48% to 76% based on self-reported figures.
- Supports major development integrations including Claude Code, OpenClaw, and CodeBuddy.
Tencent Cloud has officially released TencentDB Agent Memory v2.0 as an open-source project under the MIT license, enabling developers to easily self-host the system. Deployment can be executed swiftly via three Docker images published on Docker Hub using a single command, with multi-architecture builds covering both linux/amd64 and linux/arm64 environments.
The system systematically converts operational workflows into four distinct core asset types, which include:
- Chat Memory
- Skills
- Knowledge
- CodeGraph
All four asset categories register uniformly within the Memory Hub, ensuring that ownership, versioning, status, and visibility behave identically across the board. Specifically, Chat Memory utilizes a hierarchical structure where conversations are initially saved at L0 before an asynchronous pipeline refines them into L1 Atom, L2 Scenario, and L3 Core/Persona tiers.
Stock photo for illustration only, not from the actual event
Retrieval operations are similarly layered. L2 and L3 tiers supply rapid context bootstrapping, while specific factual queries trigger a fallback mechanism to L1 and L0 utilizing BM25, vector retrieval, and RRF. Results are capped by item count, character budget, and timeout constraints to prevent memory overcrowding within the context window.
Memory management remains a pivotal bottleneck for AI coding agents as they scale to handle complex, multi-file software repositories. By open-sourcing TencentDB Agent Memory v2.0 with granular access control lists and a built-in Cost Guard feature, Tencent Cloud provides engineering teams with a robust infrastructure framework to maintain persistent agent context while keeping enterprise data security intact.
To govern security, Memory Hub implements fixed binding combined with Access Control Lists (ACL). It narrows permissions by team, user, agent, and visibility settings prior to executing any data retrieval. Visibility modes include private, team, and restricted, with new Chat Memories and Skills defaulting strictly to private to ensure that any sharing requires an explicit action from the owner.
The version 2.0.0 release introduces several powerful capabilities, featuring skill forced archiving, scheduled CodeGraph repository synchronization, system-admin asset management, English and Chinese panel toggling, and a Cost Guard utility that assigns cheaper models to designated agents. Furthermore, the release notes report that PersonaMem accuracy climbs from 48% to 76%, representing a 59% relative improvement, though this metric is self-reported without independent verification.
Source: MarkTechPost
Found something wrong in this article? Report an issue with this article
Comments
Leave a Comment