MemOS

MemOS:

Persistent Local Memory for OpenClaw Agents

What Is This

MemOS is a local memory skill for the Happycapy Skills platform, designed specifically for OpenClaw-based agents. It provides a persistent, on-device memory layer backed by SQLite. All agent data is stored in a local SQLite database located at ~/.openclaw/memos-local/memos.db. This approach ensures that memory is retained across agent restarts and updates, enabling agents to recall, summarize, and evolve their knowledge over time. MemOS is identified by the skill ID public and is categorized under productivity.

The skill is triggered by user intents such as "install memos," "setup memory," "openclaw memory," or "configure memory." It is tailored to users who want reliable, private, and persistent storage for their agent’s memory, without the need for cloud-based solutions.

Why Use It

Many AI agents and assistant frameworks lack robust, persistent memory systems. Without persistent local storage, AI agents cannot remember previous conversations or tasks, severely limiting their usefulness and ability to personalize interactions. MemOS solves this problem by:

• Providing reliable, on-disk memory that persists between agent restarts
• Allowing agents to store, retrieve, and summarize user interactions and task data
• Enabling richer, context-aware agent behavior
• Offering privacy and control by keeping all data on the user’s device
• Requiring minimal configuration and no external dependencies beyond SQLite

These features make MemOS a foundational component for any OpenClaw agent that needs to learn from, adapt to, or recall user interactions across sessions.

How to Use It

Installation

MemOS can be installed via the Happycapy Skills platform using natural language commands. For example:

install memos

or

setup memory

Alternatively, you can enable it through the Happycapy Skills interface by searching for "MemOS" under the productivity category.

Configuration

By default, MemOS installs its SQLite database at:

~/.openclaw/memos-local/memos.db

No further configuration is typically required. However, ensure that the user account running the agent has read and write permissions to this directory.

Example:

Storing and Retrieving Data

Interacting with MemOS from an OpenClaw agent generally involves using the platform's memory API, which MemOS transparently implements on top of SQLite.

Storing a memory record:

from openclaw.memory import Memory

memory = Memory()  # MemOS-backed instance
memory.save("user_intent", {"text": "Remind me to call Alice"})

Retrieving memory records:

recent_memories = memory.query("remind")
for mem in recent_memories:
    print(mem["text"])

Upgrading MemOS

To upgrade MemOS or update its memory plugin, use commands such as:

upgrade MemOS

or

update memory plugin

The skill will handle migration and data preservation automatically.

When to Use It

MemOS is best used in scenarios where agents require durable memory for:

• Summarizing and recalling past user requests
• Evolving their skillset based on historical data
• Maintaining context across long-running tasks or conversations
• Enabling advanced features such as memory viewers or task summarization tools

It is especially valuable in privacy-conscious environments, as data never leaves the user’s local machine.

Common triggers to invoke or configure MemOS include:

• "memory not working"
• "enable memory"
• "add memory plugin"
• "memos onboarding"

These scenarios are well-supported by MemOS’s persistent, local-first design.

Important Notes

• Storage Location: All data is stored locally in ~/.openclaw/memos-local/memos.db using SQLite. Users are responsible for backing up or migrating this file if needed.
• Compatibility: MemOS is designed for OpenClaw agents. Non-OpenClaw frameworks may require adaptation or will not be compatible.
• Privacy: Since all data resides on the local device, MemOS does not transmit any information externally. Users retain full control over their data.
• Skill Evolution: MemOS is central to enabling agents to evolve by summarizing previous tasks and learning from them.
• No Cloud Dependency: MemOS works offline and does not require any cloud connection.
• Troubleshooting: If memory is "not working," check file permissions on the database path and ensure the skill is enabled.

For more details and source code, visit the MemOS GitHub repository.

MemOS provides a simple, secure, and effective way to give OpenClaw agents the memory they need to be truly intelligent and context-aware.