Openclaw Memories for Openclaw

A sophisticated memory system for AI agents combining meta-learning design evolution, structured fact extraction, and local indexing.

arosstale

v2.0.1

Feb 25, 2026

Install

Information

Skill.md

Install & Download

1. ClawHub CLI

The fastest way to install a skill directly from the registry.


                  npx clawhub@latest install openclaw-memory-2

2. Manual Installation

Copy the skill folder to one of these locations

Global

~/.openclaw/skills/

Workspace

<project>/skills/

Priority: Workspace > Local > Bundled

3. Prompt Installation

Copy this prompt to OpenClaw to install it automatically.


                  Help me install openclaw-memory-2 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 Openclaw Memories?

Openclaw Memories is a robust memory layer designed for AI agents, providing three core capabilities to handle information retention and retrieval. It leverages ALMA (Algorithm Learning via Meta-learning Agents) to evolve memory designs, uses the Observer component for remote LLM fact extraction, and employs an Indexer for offline full-text search. This setup ensures that Openclaw Skills can maintain context, track entities, and refine their own memory architecture over time through mutation and evaluation.

By integrating these components, developers can build agents that not only remember past interactions but also organize those memories into structured formats like biographical facts, world knowledge, and opinions with varying confidence scores. The system is designed for high performance with zero runtime dependencies and a lightweight footprint of under 600 lines of code.

Openclaw Memories Use Cases

Evolving and optimizing memory architectures for complex agentic workflows using meta-learning.
Extracting structured facts from long-form conversation histories using remote LLM APIs.
Implementing offline search over local Markdown-based memory repositories for better privacy and speed.
Tracking entity-specific information and user opinions over time within Openclaw Skills.
Managing daily logs and entity summaries to maintain long-term context across multiple sessions.

How Openclaw Memories Works

The ALMA component proposes and evaluates memory designs using gaussian mutation and simulated annealing to find the most efficient architecture.
The Observer component analyzes conversation history via remote LLM APIs (OpenAI, Anthropic, or Gemini) to identify entities and extract structured facts.
Extracted facts are categorized by kind (world fact, biographical, etc.), priority, and confidence, then returned as structured data.
The Indexer chunks local Markdown files, including core facts, daily logs, and entity summaries, to create a searchable index.
Agents query the Indexer to retrieve relevant context during active sessions, while the Observer continues to feed new facts into the system.

Openclaw Memories Setup

To add this memory system to your Openclaw Skills, install the package via npm:

npm install @artale/openclaw-memory

To use the fact extraction features, configure your environment variables for the Observer component:

export OPENAI_API_KEY='your-key-here'
# OR
export ANTHROPIC_API_KEY='your-key-here'

ALMA and the Indexer operate entirely offline and do not require API keys or network access.

Openclaw Memories Data Schema & Taxonomy

The system organizes data within your workspace using a specific Markdown-based taxonomy for clear organization and searchability:

File Path	Purpose
`MEMORY.md`	Stores core facts and high-priority global knowledge.
`memory/YYYY-MM-DD.md`	Contains daily logs of interactions and specific event captures.
`bank/entities/*.md`	Individual summaries and profiles for mentioned people or places.
`bank/opinions.md`	Tracks specific beliefs and opinions with confidence ratings (0.0-1.0).

Openclaw Memories Advanced Features

Self-evolving memory designs using ALMA meta-learning algorithms to optimize retrieval and storage.
Structured fact extraction with metadata including confidence scores and entity mapping for Openclaw Skills.
Offline-first indexing and full-text search over local workspace files without requiring external database dependencies.
Graceful degradation logic that returns empty arrays if remote LLM APIs are unavailable, preventing agent crashes.
Support for multi-provider LLM integrations (OpenAI, Anthropic, Gemini) for the Observer component.