- CollectionMemory & State
- Classificationagent-native
- LicenseMIT (LICENSE file; GitHub SPDX shows NOASSERTION)
- MCPUnavailable
- Agent SkillUnavailable
- Interfaceshttp
- Verified2026-09-30
What it does
TencentDB Agent Memory is a layered memory system for long-horizon agents. Short-term memory offloads verbose tool logs and keeps a compact Mermaid symbol graph with node_id drill-down to raw refs/*.md. Long-term memory builds L0 conversation → L1 atoms → L2 scenarios → L3 persona instead of a flat vector pile. GitHub description also frames a team hub of Chat Memory, Skill, LLM-Wiki, and Code-Graph assets. Official README quotes relative OpenClaw benchmark deltas (WideSearch, SWE-bench, AA-LCR, PersonaMem); treat those as upstream-reported, not independently re-measured here.
Agent-facing interfaces
- Interaction pattern: OpenClaw plugin or Hermes gateway plugin
- Catalog protocols: http
- MCP: ⚠️ Not published as a standalone MCP server
- Agent Skill: ⚠️ Not published as npx skills add
- Onboarding type:
other
How agents onboard
Interaction pattern: OpenClaw plugin or Hermes gateway plugin
openclaw plugins install @tencentdb-agent-memory/memory-tencentdb
openclaw gateway restart
See the full onboarding section in the dossier.
Protocol surface
| Interface | Detail |
|---|---|
| npm plugin | @tencentdb-agent-memory/memory-tencentdb |
| OpenClaw | openclaw plugins install … |
| Hermes Gateway | HTTP capture/search/recall (example :8420) |
| Local store | SQLite + sqlite-vec by default |
Verification
Last verified 2026-09-30.
Sources:
- https://github.com/TencentCloud/TencentDB-Agent-Memory/releases/tag/v2.0.1