A zero-dependency local memory plugin powered by SQLite and sqlite-vec to bring powerful persistence and hybrid memory recall to your AI workflows.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install local-tencentdb-agent-memory-flash
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install local-tencentdb-agent-memory-flash using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is TencentDB Agent Memory?
TencentDB Agent Memory is a high-performance, open-source conversational memory plugin created by the TencentDB team. Designed to run locally with zero external dependencies or initial configurations, it seamlessly adds persistent memory capabilities to your local AI agents. By leveraging SQLite and sqlite-vec directly, it bypasses complex database setups while ensuring quick read and write cycles.
This extension is an essential addition for developers exploring Openclaw Skills, providing an efficient way to persist preferences, recall past context, and refine tool-calling interactions without relying on cloud-hosted vector databases.
TencentDB Agent Memory Use Cases
- Personalizing Assistant Personas: Retain explicit user instructions, names, and environment preferences across completely new sessions without cluttering the system prompt.
- Long-Term Context Protection: Maintain context over extended debug sessions, preventing early context compression from purging important logical decisions.
- Local-First Workflows: Run completely offline AI environments where embedding calculations can be handled locally via BM25 hybrid search.
How TencentDB Agent Memory Works
- Memory Extraction: Evaluates conversational turns and automatically extracts key factual declarations, deduplicating repetitive statements.
- Pipeline Triggering: Runs background refinery tasks every N conversation cycles or after a short idle timeout to condense dialogue fragments.
- Hybrid Recall Retrieval: Combines keyword-based search (BM25) and local vector embeddings to retrieve relevant background context on demand.
- Context Injection: Intelligently updates active agent states with highly contextual historical facts, preventing memory recall degradation during interaction.
TencentDB Agent Memory Setup
1. Installation Check
Ensure the plugin is not already installed before starting:
npm list -g tencentdb-agent-memory 2>&1
# Or search extension list directly:
openclaw plugins list 2>&1 | findstr memory-tencentdb
2. Install Plugin
If not found, execute the installation:
openclaw plugins install @tencentdb-agent-memory/memory-tencentdb
3. Apply Patch for Tool Calls (Optional)
To ensure tool calling executions are correctly memorized, run the following patch:
cd ~/.openclaw/extensions/memory-tencentdb
bash scripts/openclaw-after-tool-call-messages.patch.sh
4. Restart the Gateway
Apply changes by restarting:
openclaw gateway restart
TencentDB Agent Memory Data Schema & Taxonomy
Configuration Architecture (~/.openclaw/openclaw.json)
The system dynamically manages configurations under the memory-tencentdb namespace:
| Category | Parameter | Default / Recommended | Description |
|---|---|---|---|
| Extraction | enableDedup |
true |
Prevents writing duplicate memories. |
| Extraction | maxMemoriesPerSession |
30 |
Upper threshold for single session storage. |
| Pipeline | everyNConversations |
3 |
Number of cycles before triggering extraction. |
| Pipeline | l1IdleTimeoutSeconds |
60 |
Delay threshold before local extraction triggers. |
| Recall | strategy |
"hybrid" |
Dual vector & keyword-based recall mechanism. |
| Recall | scoreThreshold |
0.25 |
Low thresholds maximize background retrieval accuracy. |
| Embedding | provider |
"none" |
Uses native SQLite BM25 when set to none. |
| Clean | keepDays |
30 |
Lifespan of raw L0 log storage. |
Workspace Directory Configuration
Developers can enrich the memory context by manually organizing personality details in ~/.openclaw/workspace/:
- SOUL.md: Identifies AI assistant identity rules, tone preferences, and core constraints.
- USER.md: Documents user background, workspace settings, and current code stacks.
TencentDB Agent Memory Advanced Features
- Custom Embedding Providers: Swap from the default local BM25 engine to external cloud providers like OpenAI embeddings in high-performance environments.
- Automated Pipeline Cleanup: Self-cleaning mechanisms purge raw logging history older than 30 days to optimize disk usage.
- Deep Tool Integration: Patches active gateway files to log tool execution states, broadening memory tracking to include successful system edits.
- Smart Session Pruning: Advanced context protection parameters ensure local active caches preserve a soft-threshold buffer size during long-running tasks.
SKILL.md
Loading
Related Openclaw Skills
Polymarket Trade Skills
A comprehensive, 13-skill AI-driven toolkit designed for end-to-end Polymarket prediction market analysis, execution, and portfolio management in both English and Chinese.
Bohrium Job Management
A powerful workflow automation skill for managing high-performance scientific compute jobs on the Bohrium platform directly via the bohr CLI or API.
App Feature Planner Skill
A structured AI planning agent skill designed to guide users through safe, template-driven product and application feature planning.
McDonald's MCP Ordering
An AI-ready Model Context Protocol (MCP) tool integration that enables autonomous food ordering, delivery, and menu querying for McDonald's China using Openclaw Skills.
Echo Gate
A local-first security gateway and control layer for registering, managing, and auditing agent-callable tools.
Bohrium Container Image Management
A specialized tool for managing scientific and technical Docker images, pulling custom images, and executing remote Dockerfile builds on Bohrium using Openclaw Skills.
Featured*
