TencentDB-Agent-Memory: Share Reusable Memory Across AI Agents

Overview

TencentDB Agent Memory is a self-hostable memory hub and proxy for AI agent teams. It addresses repeated context gathering by turning conversations, documents, and code into assets that can be managed and equipped for different agents.

Its main distinction from a basic chat-history store or RAG setup is the team layer: assets have owners, visibility controls, and agent bindings. It makes most sense when multiple agents or people need to reuse project knowledge, rather than for a single agent with a small, static prompt.

Key Features

• Distills conversations into layered Chat Memory, including facts, preferences, scenarios, and longer-term profiles.
• Extracts reusable Skills from conversations and tool usage, with versions, resources, and validation rules.
• Builds a linked Wiki from documents and indexes code into CodeGraph for symbol, call-relationship, and impact searches.
• Provides a Memory Hub to review, manage, and assign assets to agents.
• Supports private, team, and restricted access, including user, role, and agent ACLs.
• Offers a Proxy integration for supported agent clients, with the README describing a shared server and no required plugin, hook, or MCP server.
• Supports importing existing repositories, documents, and conversation sessions to seed a new team’s memory.

Use Cases

• A software team can give coding and review agents access to project documentation and code relationships before making changes.
• A team using agents across repeated tasks can turn proven procedures into Skills and assign them to the agents that need them.
• Teams onboarding new agents can import existing code, documents, and conversation sessions to reduce repeated discovery work.
• Organizations with separate personal and shared context can manage asset visibility and agent access centrally.

Getting Started

The README describes starting the core, hub, and proxy together with Docker-based deployment scripts:

git clone https://github.com/TencentCloud/TencentDB-Agent-Memory.git
cd TencentDB-Agent-Memory/deploy/global-images
cp .env.example .env
./start-all.sh

Configure the LLM parameters in .env before starting. See the README and installation guide for setup and client-specific instructions.

Alternatives

• Cabinet: Cabinet combines a private knowledge base with agent teams and workflow automation, while TencentDB-Agent-Memory focuses on reusable context and skills shared across agents.

Considerations

• Deployment requires configuring two sets of LLM parameters, for the memory group and proxy group. The README specifies Node.js 22.16 or newer and documents a Docker-based deployment path.
• MongoDB storage is experimental and disabled by default.
• Wiki and CodeGraph processing is asynchronous, so imported assets may take time to become ready. CodeGraph currently prioritizes public HTTPS repositories; private repository and SSH support is still being refined.
• Manual asset binding is supported, while fully automated memory routing is still under iteration.
• GitHub reports the license as NOASSERTION, while the README displays an MIT license badge. Verify the repository’s license files and terms before adoption.