A comprehensive AI memory framework that categorizes user, feedback, project, and reference data using persistent Markdown storage and MiniMax LLM-driven consolidation.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install memory-system-pro-v2
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 memory-system-pro-v2 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 Memory System?
Memory System is a sophisticated long-term memory layer designed for AI coding agents within the Openclaw Skills ecosystem. Unlike transient session memory, this skill utilizes a local file-based storage approach using pure Markdown, making your agent's knowledge base transparent, editable, and portable. It leverages semantic search via Ollama and vector embeddings to ensure the most relevant context is always available to the LLM.
The core of the system is its ability to learn from interactions. By categorizing information into four distinct types—User, Feedback, Project, and Reference—the agent maintains a nuanced understanding of your preferences, technical standards, and ongoing project goals. The integration of MiniMax LLM ensures that memory stays clean and relevant through periodic background processing.
Memory System Use Cases
- Automatically capturing user coding preferences and technical background for personalized responses.
- Recording critical feedback or corrections to prevent the AI from repeating the same mistakes.
- Managing complex project context, including goals, timelines, and architectural decisions across multiple sessions.
- Maintaining a structured index of external resources like documentation links and issue trackers.
How Memory System Works
- Context Capture: The skill monitors interactions and writes significant information to categorized Markdown files using the memory_write tool.
- Semantic Retrieval: When a new session starts, the system uses semantic vector search to automatically load relevant memories into the current context.
- Bidirectional Feedback: Users can provide positive or negative feedback, which is specifically stored to refine the agent's future behavior and rule application.
- AutoDream Consolidation: On a scheduled basis or manual trigger, the system scans all memory files and session history.
- LLM Processing: Using the MiniMax LLM, the system identifies outdated info, merges duplicates, and updates the central MEMORY.md index.
- Intelligent Flushing: When context windows approach their limits, the system automatically flushes memory to disk to prevent data loss.
Memory System Setup
To install the memory system as part of your Openclaw Skills library, use the following CLI command:
clawhub install memory-system
Configure your environment variables for the consolidation features:
{
"skills": {
"memory-system": {
"memoryDir": "~/.openclaw/workspace/memory",
"autoDream": {
"enabled": true,
"apiKey": "env:MINIMAX_CODING_API_KEY"
}
}
}
}
Memory System Data Schema & Taxonomy
The system organizes data in a structured directory tree using Markdown files for maximum readability:
| File/Folder | Description |
|---|---|
MEMORY.md |
The central index file for all stored memories |
user/ |
Stores profiles, roles, and preferences |
feedback/ |
Contains positive/negative corrections and application rules |
project/ |
Tracks progress, goals, and project-specific decisions |
reference/ |
Links to external systems like Jira, Linear, or Confluence |
sessions/ |
Raw session history stored in JSON format |
Memory System Advanced Features
- AutoDream: Automated background maintenance that performs self-cleanup and indexing every 24 hours or after 3 new sessions.
- Vector-Search Hybrid: Combines high-performance vector embeddings with keyword fallback for reliable retrieval.
- Heartbeat Integration: Deeply hooks into the Openclaw heartbeat for silent, scheduled maintenance tasks.
- Smart Thresholding: Configurable token thresholds (default 300,000) for automated memory persistence (Flush).
SKILL.md
Loading
Related Openclaw Skills
PDF Reader Assistant
An AI-powered PDF assistant for extracting content, generating smart summaries, and converting documents into clean Markdown format.
News Aggregator
An automated intelligence gathering tool that searches, filters, and summarizes global technology and military news from authoritative sources.
Quiz Generator AI Skill
A comprehensive AI-driven tool for generating automated quizzes, personality tests, and professional surveys with integrated scoring logic.
Douyin Search & Comment Skill
An automated workflow for searching viral Douyin videos and posting human-like comments using high-performance browser automation.
bug-fixer
An autonomous AI agent skill designed to streamline the identification, diagnosis, and resolution of software bugs.
Bug Fixer
An autonomous agent skill for the systematic diagnosis, repair, and verification of software bugs and system errors.
Featured*
