Technical Guide

TencentDB-Agent-Memory

TencentDB-Agent-Memory is built for agent teams that need conversations, documents, and code to remain reusable after a single run ends.

The repository is not trying to extend one chat transcript. It is trying to keep project context available for the next handoff, review cycle, or agent run.

Open install guide
English +1
English简体中文
TencentCloud/TencentDB-Agent-Memory
AI Toolv2.0.0
Stars
20,272
Primary language
TypeScript
License
MIT
Last updated
8/12/2026

Overview

TencentDB-Agent-Memory narrows memory scope through roles, visibility, and bindings before retrieval begins.

The model starts with Team, User, Agent, visibility, and fixed bindings, so scope is narrowed before retrieval even begins.

Documents move into Wiki assets and repositories into CodeGraph assets, which preserves link structure, symbols, and call relationships instead of flattening everything into plain text.

That makes rollout less about adding one more memory store and more about validating ingest paths, asset readiness, and access boundaries.

Best for

Best fit for teams that need project context to survive reviews, handoffs, and repeated work.

  • Choose it when new agents or teammates need to start from imported repositories, documents, and prior runs instead of an empty session.
  • It becomes more valuable when the same context keeps reappearing in reviews, debugging, releases, or handoffs.

Features

TencentDB-Agent-Memory centers its feature set on layered memory, wiki assets, code graph analysis, and reusable skills.

  • Conversation history is distilled into multiple memory levels instead of remaining one raw transcript.
  • Document imports become Wiki assets and repository imports become CodeGraph assets, so structure survives alongside the text itself.
  • Skill assets are versioned workflow units with resource files, trigger rules, execution steps, and validation boundaries.

Not for

Less suitable for short-lived chat work or workflows that never plan to maintain shared memory assets.

  • Short-lived chat assistance does not need this much structure.
  • Workflows that never import repositories, documents, or prior sessions will leave much of the asset model unused.