Reduce repetitive learning for AI agent teams and share team experience as assets
TencentDB Agent Memory is a centralized memory hub for AI agent teams. It converts conversations, documents, and code into four reusable assets: chat memory, skills, wiki, and code graph. It helps teams produce consistent results based on shared memory without training each agent separately.
v2.0.0 Release and Team Memory Beta
The current latest version is v2.0.0, and the team memory beta feature is evolving rapidly. You can start exploring within minutes of installation. Existing v1.x or v0.x users can migrate their data to v2.0.0 or later using the data migration tool. New installations can skip this process and start immediately.
Accumulating Experience Through Four Memory Assets
This tool aims to structure information so that the next agent does not have to reinvent the wheel, rather than simply storing conversation logs.
- Chat Memory Stores facts, preferences, and decisions extracted from conversations, refined and stored hierarchically from L0 to L3. Agents are automatically assigned their own memory upon creation.
- Wiki Converts documents and files into structured pages and link graphs. Allows agents to quickly find necessary information without reading all file lists.
- Code Graph Indexes code symbols, files, call relationships, and impact paths. Enables checking callers and callees and analyzing impact scope before modifying code.
Zero-Code Integration via Proxy
Designed so that all agents share the same memory server. You only need to change the agent's base URL to a proxy, without plugins, hooks, or MCP servers. Supports various frameworks such as DeepSeek Harness, Claude Code, Codex, and CodeBuddy. Unsupported agents can be adapted using the general integration guide, and PRs for adding native support are welcome.
Cold Start and Team Management Panel
Existing codebases, documents, and conversation sessions can be imported and processed automatically, reducing the cost of initial learning for new agent teams. The memory hub panel allows creating teams and agents, and managing asset ownership, version, status, visibility, usage count, and agent binding in one place. Three visibility levels—personal, team, and ACL-based access control—allow precise adjustment of information sharing scope.
Pre-Application Checks and Limitations
This tool is suitable for reducing repetitive context explanations and document re-reading, and for sharing validated workflows. However, wiki and code graph are generated asynchronously, so you must wait for processing time until they reach the ready state. Code graph currently prioritizes public HTTPS repositories, while support for private repositories and SSH authentication is still under improvement. Additionally, fully automated memory routing is still in iterative development, requiring manual asset binding. It is released under the MIT license and continues to evolve through community contributions.
