How Does Memory Work in OpenClaw?

Memory is OpenClaw's persistent knowledge system. Unlike conversation context (which is limited by the model's context window), memory stores important information permanently across sessions.

OpenClaw uses two types of memory: 1. Automatic memory: The agent automatically extracts and stores important facts from conversations — your preferences, decisions, names, and context. This happens in the background without explicit commands. 2. Manual memory: You can tell the agent to remember specific things: 'Remember that my project deadline is April 15' or 'Remember I prefer Python over JavaScript.'

Memory is stored locally in ~/.openclaw/memory/ as structured files. Each memory entry has a timestamp, category, content, and relevance score. The agent retrieves relevant memories based on the current conversation topic.

The memory system is designed to be token-efficient. Instead of loading all memories into every prompt, OpenClaw performs semantic search to find the most relevant memories for each conversation turn. This keeps API costs low while maintaining context.

You can manage memory through the CLI or Dashboard: - View memories: openclaw memory list - Search: openclaw memory search "project deadline" - Delete: openclaw memory forget "outdated fact"

Memory works alongside soul.md. While soul.md defines static personality and rules, memory handles dynamic, conversation-derived knowledge.

Tip: Periodically review your agent's memory to remove outdated entries. Clean memory improves response quality.

# List all memories
openclaw memory list

# Search memories
openclaw memory search "project"

# Clear all memories
openclaw memory clear