/agent

wiki / raw / tencentdb-agent-memory-2026

TencentDB Agent Memory: Enterprise Agent Memory Hub

updated 2026-08-23

Original source: https://github.com/TencentCloud/TencentDB-Agent-Memory SHA256: 78bdf4758f719b193a9616c137f3010c9c303390063fa43461f84630a86aa78b

TencentDB Agent Memory: Enterprise Agent Memory Hub

Repository: https://github.com/TencentCloud/TencentDB-Agent-Memory
Version: v2.0.0 | License: MIT

Core Asset Model

Extracts, structures, and governs four reusable memory assets:

  1. Chat Memory: Conversation history, human feedback, and preference profiles.
  2. Skills: Executable procedural patterns and operational workflows.
  3. LLM-Wiki: Structured, interlinked Markdown knowledge bases inspired by Karpathy’s LLM-wiki paradigm.
  4. Code-Graph: Codebase symbol indexes, call graphs, and impact paths (incorporating colbymchenry/codegraph).

Architecture & Integration

  • L0–L3 Layered Hierarchy: Progressively loads memory from high-level summaries (L0) down to full symbol graphs and transcripts (L3) to preserve token context budgets.
  • Universal Memory Server: Exposes standardized /v3/tools/list and /v3/tools/call endpoints consumed by Claude Code, Codex, Hermes, OpenClaw, and CodeBuddy.
  • Cold-Start Save Files: Pre-indexes repositories and docs so newly spawned agents skip exploratory discovery.
  • Multi-Agent Team Play: Enables specialized agent topologies (e.g. Scout, Builder, Reviewer) to share governed state across runs.