Verified project record

TencentCloud/TencentDB-Agent-Memory

tencentcloud/tencentdb-agent-memory provides hierarchical long-term memory and symbolic short-term memory compression to reduce token usage for AI agents. Developers can run it locally with a SQLite backend, though it requires an OpenAI-compatible LLM and carries a planned status for cross-agent portability.

Project overview

Addresses agent context loss and token waste using a semantic pyramid and symbolic short-term memory compression.

Project type: AI Agent · RAG
Use cases: Automation
Deployment: Refer to project documentation
License: License pending

Best for

Developers and AI engineers building agents that require persistent context and reduced token consumption during long-horizon tasks.
Teams seeking a local-first memory backend using SQLite without initial configuration.

Key capabilities

Builds a semantic pyramid from L0 Conversations to L3 Personas instead of relying on flat vector piles.
Offloads heavy tool logs to external files and condenses task state into a lightweight Mermaid canvas.
Defaults to a local SQLite and sqlite-vec backend.
Supports both keyword and semantic recall using BM25, vector search, and RRF fusion.
Exposes capture, search, and recall HTTP endpoints via a sidecar Gateway service.
Cross-agent, cross-framework, and cross-device memory import, export, and live migration.

Limitations and risks

The OpenClaw integration requires version 2026.3.13 or later.
The Gateway may issue a warning if exposed without an API key configured.
Requires an OpenAI-compatible LLM and embedding model for extraction, meaning fully local execution without external services is limited.

Getting started

Setup involves medium difficulty due to dependency installation and patch scripts for certain configurations. The first success path requires installing the plugin via OpenClaw, configuring the provider, and verifying the Gateway.

Evidence and sources

GitHub project description: TencentDB Agent Memory delivers fully local long-term memory for AI Agents via a 4-tier progressive pipeline, with zero external API dependencies.
README: openclaw plugins install @tencentdb-agent-memory/memory-tencentdb
README: In practice, we constantly re-explain the same SOPs, project background, tool conventions, and output formats to the Agent. Such information should not require repetition
README: When integrated with OpenClaw, it cuts token usage by up to 61.38%, improves pass rate by 51.52% (relative), and raises PersonaMem accuracy from 48% to 76%.
README: 3. Production-Ready Engineering: Not a Demo

Reviewed at: 2026-07-18T04:13:09Z