Making OpenClaw Actually Remember Things

Overview

Ray discusses how to optimize OpenClaw's long-term memory using QMD, a free, open-source mini CLI search engine for markdown files. He shares his memory optimizer skill, which includes a dream cycle and morning brief to clean up and organize memories, and he explains the three-tier memory system to reduce token usage.

Key takeaways

OpenClaw's memory system relies on markdown files for storing information, including daily logs and agent instructions.
QMD, a free and open-source CLI search engine, can be integrated with OpenClaw to enhance memory search and retrieval.
Ray's memory optimizer skill includes a dream cycle and morning brief to clean up and organize memories, improving agent efficiency.
The three-tier memory system (Tier 1, Tier 2, Tier 3) allows for progressive disclosure, optimizing token usage by loading only necessary information.
Exa.ai is Ray's primary search engine for OpenClaw, providing filtered data sets and reducing the risk of prompt injection.
Updating OpenClaw to the latest version (2026.2.15) and enabling QMD can significantly improve memory management and agent performance.

Chapters

0:00 OpenClaw's Memory System and OpenAI Integration

OpenClaw is now under OpenAI's direction, with Pete Steinberger joining the OpenAI team.
The memory system in OpenClaw uses plain markdown files stored in a workspace, including daily logs and an agents.md file that steers the language model.
OpenClaw has an automatic memory flush that triggers a silent agentic turn when a session is close to auto-completion, writing any lasting notes before compaction.

5:57 QMD Integration for Enhanced Memory Search

QMD is a free, open-source mini CLI search engine for markdown documentation that can be integrated with OpenClaw for improved memory search.
QMD allows users to search through knowledge bases, media notes, and track current states, providing a fast and ranked information retrieval system.
To enable QMD, users should update OpenClaw to version 2026.2.15 and instruct their agent to enable the memory backend to be QMD.

20:22 Memory Optimizer Skill and Progressive Disclosure

Ray created a memory optimizer skill with a dream cycle and morning brief to clean up and organize memories, which includes a script to check file sizes and provide recommendations.
The dream cycle reviews daily files and connects the dots for recent days, while the morning brief encourages conversation with the agent.
Ray describes a three-tier memory system: Tier 1 (expensive, always loaded), Tier 2 (searchable, on-demand via QMD), and Tier 3 (full file read, on-demand), enabling progressive disclosure for memory management.

Summary, takeaways, and chapters were generated by AI from the video's transcript and may contain errors. The video belongs to its creator, Ray Fernando.