A high-traceability memory continuity system for AI agents that uses structured Markdown files as a persistent source of truth.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install duru-memory
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 duru-memory 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 Duru Memory?
Duru Memory is a sophisticated memory management framework designed to provide AI agents with long-term continuity through local Markdown files. By treating plain text as the source of truth, it ensures that project states, decisions, and handoff notes remain human-readable and easily auditable. It serves as a specialized extension to the standard Openclaw Skills ecosystem, offering a robust structure for organizing daily logs, core rules, and project-specific knowledge.
This skill bridges the gap between ephemeral session data and permanent knowledge. While it leverages built-in retrieval tools for speed, Duru Memory provides the underlying architecture and maintenance scripts necessary to keep an agent's memory clean, tagged, and relevant over months of operation. It is particularly effective for developers who need their AI collaborators to maintain a deep understanding of evolving project contexts without data loss between sessions.
Duru Memory Use Cases
- Maintaining state continuity across multiple AI agent chat sessions.
- Creating a permanent, searchable audit trail of technical decisions and project blockers.
- Implementing standardized handoff protocols for multi-agent workflows.
- Managing complex project memories where high traceability and human-readable logs are required.
- Debugging agent recall by using deterministic file-level search instead of opaque vector lookups.
How Duru Memory Works
- The agent initializes the session by loading core rules and current state files from the defined directory structure.
- Recent daily logs (typically the last 48 hours) are ingested to provide immediate context of ongoing tasks.
- During operation, the system prefers built-in memory search for speed but can trigger deterministic local scripts for precise file-level control.
- Every meaningful change in task status or project direction is recorded as a state diff in the changelog.
- Upon session termination, a closing protocol runs to summarize activities, append daily logs, and ensure state freshness for the next interaction.
Duru Memory Setup
To integrate Duru Memory into your environment, ensure your workspace follows the standard directory contract and install the necessary utility scripts.
# Create the required directory structure
mkdir -p memory/{CORE,daily,projects,people,concepts,handoff,archive/raw}
# Initialize core memory files
touch memory/CORE/hard-rules.md memory/CORE/current-state.md
# Run the session start script to load context
./scripts/session-start.sh
Ensure your config.yaml is configured with the correct local model paths if you intend to use the optional semantic search features with Openclaw Skills.
Duru Memory Data Schema & Taxonomy
The system organizes memory into a strict hierarchy to ensure fast retrieval and clear categorization:
| Directory | Content Type | Retention Policy |
|---|---|---|
CORE/ |
Hard rules, current state, and state changelogs | Permanent |
daily/ |
Incremental logs of daily activities | Compacted weekly |
projects/ |
Deep-dive context for specific repositories | Active until archive |
people/ |
User preferences and collaborator profiles | Permanent |
handoff/ |
Context packets for session transitions | Temporary |
Each entry supports metadata attributes such as status (active/superseded), polarity (positive/negative), and confidence levels to assist in weighted retrieval.
Duru Memory Advanced Features
- Hybrid Retrieval Engine: Combines keyword-based deterministic search with experimental local semantic retrieval via SQLite-vec.
- Automated Maintenance: Includes scripts for weekly compaction (summarizing logs) and monthly 'forgetting' of stale data while preserving critical negative pitfalls.
- Auto-Tagging System: Utilizes local LLMs to automatically categorize and review memory entries for metadata consistency.
- Negative Memory Handling: Specifically tracks and surfaces 'avoided pitfalls' to prevent the agent from repeating past mistakes.
- Git Integration: Optional auto-commit hooks to ensure the memory tree is version-controlled and backed up.
SKILL.md
Loading
Related Openclaw Skills
LLM Memory Integration
A sophisticated dual-layer memory management system providing AI agents with persistent, hybrid search capabilities.
Blackboard Dark Poster Generator
An AI agent skill that generates high-contrast dark posters and tutorial covers featuring authentic handwritten marker typography and minimalist design.
Gemini LLM Connector
A high-performance interface for one-shot Q&A, summarization, and text generation powered by the SkillBoss API Hub.
Interview Pro
An all-in-one AI assistant for mock interviews, resume optimization, and industry-specific interview preparation.
LLM Memory Integration
A hybrid memory management system providing vector search and SQLite-based FTS for persistent AI agent context.
Agent Android
A local-first automation skill for controlling Android devices over a network without requiring USB connections, ADB, or root access.
Featured*
