TencentDB Agent Memory v2.0 solves a common frustration for AI‑agent teams: every new session forces you to re‑explain project context, decisions, and code details. Instead of repeatedly feeding the same information into prompts, the system turns chats, documents, code, and learned procedures into four reusable memory assets – Chat Memory, Skill, Wiki, and CodeGraph. Each asset is versioned, permissioned, and can be scoped to specific agents, so teammates can benefit from what you have already learned without seeing anything you marked private.
Deployment is straightforward. The project is MIT‑licensed and self‑hosted. Three Docker images are available on Docker Hub and can be started with a single command, supporting both linux/amd64 and linux/arm64 architectures. Default ports are documented in the install guide, and official SDKs exist for TypeScript and Python. Integration points include OpenClaw, Hermes, Claude Code, CodeBuddy, or direct SDK calls via a Memory Proxy that speaks Anthropic and OpenAI protocols.
The real advantage lies in the governance layer. Visibility settings – private, team, restricted, and agent‑specific – let you control who can read each memory asset. By default new Chat Memory and Skills are private, making sharing an explicit act. The hub uses fixed binding plus ACL: first filter by team, user, agent, and visibility, then retrieve the relevant assets. This prevents leakage while still enabling collaborative learning.
Retrieval is layered and budget‑aware. Conversations start at raw L0, then an async pipeline distills them into L1 Atom, L2 Scenario, and L3 Core/Persona. When you need fast context, L2/L3 are used; for specific facts the system falls back to BM25, vector search, and reciprocal rank fusion, all capped by item count, character budget, and timeout to avoid overwhelming the model’s context window.
Practical applications include onboarding new agents to an existing codebase, running impact analyses before refactoring, maintaining release checklists, incident runbooks, code‑review standards, and converting product documentation into agent‑readable pages. Teams report improved persona recall (self‑stated rise from 48% to 76%), though independent verification is pending and private‑repo CodeGraph features are still maturing.
In short, if your team wastes time re‑explaining context, struggles to share agent knowledge securely, or needs a lightweight, self‑hosted memory hub that works with existing LLM workflows, TencentDB Agent Memory offers a deployable, governance‑driven solution.
#AI #Product #DevOps #LLM #Agents #Cloud