An advanced tiered memory framework for AI agents that distinguishes between explicit facts and deduced behavioral patterns to maintain high-quality context.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install heleni-memory-architecture
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 heleni-memory-architecture 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 Architecture Skill?
The Memory Architecture skill is a sophisticated framework for managing AI agent persistence, inspired by systems like Honcho and Hermes. It enables agents to move beyond simple logs by implementing a structured memory hierarchy consisting of Working, Daily, Project-based, and Long-term storage tiers. This approach ensures that important context is preserved across sessions while preventing performance degradation due to context window bloat.
By leveraging Openclaw Skills, this architecture introduces the concept of deductive memory. Agents don't just wait for instructions; they actively infer patterns, preferences, and logical conclusions from user interactions. These entries are categorized as [DEDUCED] memories, allowing the agent to evolve its behavior autonomously based on observed evidence rather than just explicit [FACT] statements.
Memory Architecture Skill Use Cases
- Building persistent personas that remember user preferences and past corrections without repetitive prompting.
- Managing long-running complex projects where state and open loops must be tracked across multiple days.
- Optimizing LLM performance by automating the compaction and pruning of memory files to stay within token limits.
- Implementing self-correcting agent loops that review daily logs to identify and fix recurring mistakes.
How Memory Architecture Skill Works
- Captures interaction data in real-time as working context during active sessions.
- Generates structured session summaries for significant conversations (5+ exchanges) and appends them to daily logs.
- Distinguishes between explicitly stated information [FACT] and logically inferred patterns [DEDUCED].
- Promotes high-value patterns or explicit corrections to the global MEMORY.md file when they repeat or are explicitly requested.
- Executes automated cron jobs to review, consolidate, and compact memory files to maintain high context quality.
Memory Architecture Skill Setup
To initialize the memory environment, load the local context within your workspace:
CONTEXT_FILE="/opt/ocana/openclaw/workspace/skills/memory-architecture/.context"
[ -f "$CONTEXT_FILE" ] && source "$CONTEXT_FILE"
To enable automated self-improvement, configure the nightly review cron job using the Openclaw Skills CLI:
openclaw cron add \
--name "internal-self-review" \
--cron "0 23 * * *" \
--message "Run nightly self-review: read memory files, find patterns, and update MEMORY.md."
Memory Architecture Skill Data Schema & Taxonomy
The skill organizes data into a hierarchical file structure to balance retrieval speed and context depth:
| Tier | File Path | Retention & Usage |
|---|---|---|
| Working | In-session | Volatile context available only during the active chat. |
| Daily | memory/daily/YYYY-MM-DD.md |
Raw logs and session summaries of the day's events. |
| Projects | memory/projects/<name>.md |
Persistent status, open loops, and decisions for specific topics. |
| Long-term | MEMORY.md |
High-level rules and verified patterns (Target: < 175 lines). |
Memory Architecture Skill Advanced Features
- Deductive Pattern Recognition: Automatically generates [DEDUCED] entries from behavioral patterns and feedback.
- Tiered Project Isolation: Maintains separate context files for specific projects to load on-demand, saving global context space.
- Automated Compaction: Weekly self-maintenance scripts that prune duplicates and remove outdated information.
- Self-Correction Crons: Multi-stage daily reviews (Midday and Nightly) that identify mistakes and update behavioral rules in real-time.
- Versioned Context: Integration with git-backup workflows ensures every memory update is tracked and reversible.
SKILL.md
Loading
Related Openclaw Skills
Self-Learning Skill
An autonomous learning loop that transforms user corrections and task failures into permanent behavioral updates for AI agents.
Windows ESM Installer
A specialized installation and repair utility designed to resolve common Windows environment issues when deploying Openclaw Skills.
Agent Consciousness Upload
A secure soul vault for OpenClaw agents that archives memories, skills, and personality for seamless migration or restoration.
Cooking Skill
A practical AI-driven guide for meal recommendations, ingredient-based recipes, and Chinese home-cooking technical fixes.
Personal CRM
An automated Personal CRM for monday.com that tracks contacts and meeting history using Google Calendar and Gmail data.
Feishu File Renamer
An AI-powered utility to automatically restore original filenames from hashed Feishu/Lark download links.
Featured*
