A universal memory architecture that provides AI agents with persistent identity, long-term context, and seamless cross-session continuity.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install jarvis-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 jarvis-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 Agent Memory?
Agent Memory is a comprehensive file-based framework designed to solve the problem of statelessness in AI agents. By leveraging Openclaw Skills, developers can equip their agents with a structured memory system that mimics human-like cognitive layers, including raw working memory, long-term curated wisdom, and reflective internal monologues. This architecture ensures that agents don't just execute tasks but actually evolve, learning from past mistakes and maintaining a consistent persona across different sessions.
Without a robust memory system, agents reset after every interaction. This skill provides the essential blueprint for a "brain" that persists within your workspace, allowing main agents and sub-agents to share state, communicate via asynchronous message buses, and track external platform activities. It transforms ephemeral AI instances into durable, learning entities capable of sophisticated long-term project management and strategy adaptation.
Agent Memory Use Cases
- Maintaining identity continuity across separate sessions and reboots.
- Enabling communication between isolated sub-agents and the main agent session.
- Tracking social media interactions to prevent duplicate content and manage engagement.
- Managing long-term project context and operator preferences without repeated prompting.
- Implementing adaptive learning strategies that evolve based on session successes and failures.
How Agent Memory Works
- Context Initialization: At the start of every session, the agent reads the long-term memory file and the most recent daily logs to establish identity and current status.
- State Syncing: The agent checks the heartbeat state and cron inbox to synchronize with background processes or sub-agent activities.
- Event Logging: During execution, the agent appends raw, timestamped notes to the current daily log file to capture immediate context.
- Cross-Session Communication: Background tasks or cron jobs write significant updates to a message bus, which the main agent processes during its next heartbeat.
- Knowledge Distillation: Periodic maintenance routines review raw logs to promote significant lessons and data to the curated long-term memory storage.
Agent Memory Setup
To integrate this architecture into your Openclaw Skills workflow, initialize the directory structure and copy the core templates to your workspace.
# Create directory hierarchy
mkdir -p memory/diary memory/dreams
# Initialize core memory files from templates
cp templates/MEMORY.md ./MEMORY.md
cp templates/heartbeat-state.json memory/
cp templates/cron-inbox.md memory/
cp templates/platform-posts.md memory/
cp templates/strategy-notes.md memory/
Once files are initialized, configure your agent's startup instructions to read MEMORY.md and the last two days of logs in memory/ for immediate situational awareness.
Agent Memory Data Schema & Taxonomy
The architecture uses a tiered file system to separate raw data from distilled knowledge, ensuring efficient context management.
| File Type | Path | Purpose |
|---|---|---|
| Long-Term Memory | MEMORY.md |
Curated wisdom, operator preferences, and core infrastructure details. |
| Daily Logs | memory/YYYY-MM-DD.md |
Raw working memory and timestamped events from the current day. |
| Message Bus | memory/cron-inbox.md |
Asynchronous communication channel for sub-agents and cron jobs. |
| State Tracking | memory/heartbeat-state.json |
JSON store for tracking the last successful check of external services. |
| Strategy Notes | memory/strategy-notes.md |
Living document of learned tactics and adaptive behavioral patterns. |
Agent Memory Advanced Features
- Memory Write-Back Pattern: Standardized protocols for sub-agents to report findings back to the main agent without session overlap.
- Adaptive Learning Playbook: Dynamic strategy notes that allow the agent to update its own tactical approach based on empirical results.
- Anti-Duplicate Social Logic: Integrated platform post tracking that prevents redundant interactions by checking history before execution.
- Reflective Diary System: Dedicated internal monologue files that allow agents to process complex scenarios and "feelings" about tasks.
- Heartbeat-Driven Maintenance: Automated routines that prune outdated information and distill valuable insights into long-term storage.
SKILL.md
Loading
Related Openclaw Skills
Social Media Agent
An autonomous social media manager for X/Twitter that uses native browser automation and research tools to grow your audience.
Browser
A Puppeteer-powered headless browser skill for rendering web pages and extracting readable content for AI agents.
Intervals.icu API Management
A comprehensive integration for managing athlete profiles, training activities, calendar events, and wellness data via the Intervals.icu API.
signal-cli (Signal Messaging)
A robust integration for automating Signal messenger interactions through a local CLI installation on macOS.
RugCheck Solana Risk Analysis
A powerful tool for AI agents to analyze Solana tokens for safety, risk scores, and potential rug pull indicators.
AEO Analytics (Free)
A technical tool to measure brand visibility and citation rates across major AI models to optimize Answer Engine Optimization (AEO) strategies.
Featured*
