A sophisticated three-tier memory architecture that mimics human cognition to provide AI agents with efficient, hierarchical, and long-term context retention.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install tiered-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 tiered-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 Tiered Memory System?
The Tiered Memory System is a professional-grade memory architecture designed to solve the context window limitations of modern LLMs within Openclaw Skills. By organizing information into Hot, Warm, and Cold tiers, it ensures that agents retain critical identity data and active project context while archiving historical facts in a searchable, tree-structured index. This system is inspired by human cognition, utilizing relevance decay and strategic forgetting to keep agent responses sharp and focused.
Unlike standard vector-based RAG, this skill employs a vectorless retrieval method based on LLM reasoning and a PageIndex-inspired tree structure. This allows for O(log n) navigation of categories, ensuring that retrieval is not only fast but also explainable and highly accurate. With its cloud-first design, the soul of your agent lives in the cloud, allowing for seamless synchronization across multiple devices and robust disaster recovery.
Tiered Memory System Use Cases
- Maintaining long-term agent identity, owner preferences, and relationship history over years of interaction.
- Organizing complex project technical architectures and decisions into a hierarchical, searchable category tree.
- Automatically distilling daily conversations into structured facts while preserving critical metadata like URLs and shell commands.
- Managing memory for multiple independent agents using scoped IDs within a single deployment.
- Validating daily notes to ensure technical information like documentation links and installation commands are not missing.
How Tiered Memory System Works
- The system ingests raw conversation text or daily notes and passes them through a distillation engine to extract core facts.
- Extracted facts are assigned an importance score and categorized within a hierarchical tree index.
- Recent facts are stored in the Warm Memory tier where they undergo a periodic decay process based on age and access frequency.
- High-importance items like core identity and active tasks are promoted to the Hot Memory tier, which is always present in the agent context.
- Aging or low-scored memories are archived into the Cold Memory tier (powered by Turso) to maintain a lean and efficient active context window.
- During retrieval, the system uses LLM reasoning to navigate the tree index and fetch only the most relevant memory nodes for the current query.
Tiered Memory System Setup
To integrate this memory architecture into your Openclaw Skills, follow these installation steps:
# Install the tiered-memory skill
clawhub install tiered-memory
# Configure your Turso database credentials for cloud sync
export TURSO_URL="your-database-url"
export TURSO_TOKEN="your-auth-token"
# Initialize the cold storage tables
python3 skills/tiered-memory/scripts/memory_cli.py cold --init --db-url $TURSO_URL --auth-token $TURSO_TOKEN
# Rebuild the initial hot state
python3 skills/tiered-memory/scripts/memory_cli.py consolidate --mode daily
Tiered Memory System Data Schema & Taxonomy
The skill manages information across several specialized files and database tables to optimize retrieval performance.
| Component | Format | Max Size | Description |
|---|---|---|---|
| Hot Memory | JSON | 5KB | Core identity, owner profile, and active context (always in-context). |
| Warm Memory | JSON | 50KB | Distilled facts from the last 30 days with decay scoring. |
| Cold Memory | SQL | Unlimited | Long-term archive stored in Turso DB for historical queries. |
| Tree Index | JSON | 2KB | Hierarchical category map for efficient navigation. |
| Metrics | JSON | N/A | Observability data tracking token savings and retrieval counts. |
Tiered Memory System Advanced Features
- LLM-powered semantic tree search that reasons about category relevance rather than relying on simple keyword similarity.
- Automated metadata extraction engine that identifies and preserves URLs, shell commands, and file paths during distillation.
- Multi-tier consolidation modes (Quick, Daily, Monthly, Full) for automated maintenance and tree pruning.
- Cloud-first synchronization that backs up critical hot state and index structure after every conversation for multi-device persistence.
- Proactive memory validation that flags incomplete technical notes or missing documentation links.
SKILL.md
Loading
Related Openclaw Skills
Solidity LSP
A comprehensive language server integration for Solidity that provides compilation, linting, and security auditing for smart contract developers.
rust-analyzer-lsp
A high-performance language server providing deep code intelligence, real-time error detection, and advanced refactoring tools for Rust development.
Pyright LSP
A high-performance Python language server providing static type checking, code intelligence, and real-time diagnostics for professional development.
Intelligent Router
A core infrastructure skill for Openclaw Skills that dynamically routes agent tasks to the most cost-effective and capable AI models.
TypeScript LSP
A powerful language server providing deep code intelligence, type checking, and refactoring for TypeScript and JavaScript projects.
WEEX Futures Trading
A high-performance AI agent skill for executing USDT-margined perpetual futures trades on the WEEX exchange with up to 125x leverage.
METADATA
Featured*
