AgentSkillsCN

mem0-usage

如何使用OpenClaw Mem0 v2增强版内存插件。

SKILL.md
--- frontmatter
name: mem0-usage
description: How to use the OpenClaw Mem0 v2 enhanced memory plugin
version: 2.0.0
metadata:
  openclaw:
    requires:
      tools: ["memory_search", "memory_store", "memory_store_raw"]

Using OpenClaw Mem0 v2 Memory

Available Tools

memory_search

Search memories by semantic similarity.

  • query (required): What to search for
  • limit: Max results (default: 5)
  • userId: Scope to specific user
  • scope: "session", "long-term", or "all" (default: "all")
  • category: Filter by category (e.g. "technical", "infrastructure", "preferences")

memory_store

Store information with LLM fact extraction.

  • text (required): Information to remember
  • userId: User scope
  • longTerm: true for user-scoped, false for session-scoped

memory_store_raw

Store verbatim text WITHOUT LLM extraction. Use for exact quotes, commands, configs, code.

  • text (required): Exact text to store
  • userId: User scope
  • category: Tag (e.g. "infrastructure", "technical")

memory_get

Retrieve a specific memory by ID.

  • memoryId (required): Memory ID

memory_list

List all memories for a user.

  • userId: User to list memories for
  • scope: "session", "long-term", or "all"

memory_forget

Delete a specific memory.

  • memoryId (required): Memory ID to delete

memory_search_log (sleep mode only)

Search conversation history logs.

  • query (required): Search text
  • dateFrom: Start date filter
  • dateTo: End date filter
  • limit: Max results

Categories

Memories are auto-categorized into:

  • identity: Name, age, location, occupation
  • preferences: Likes, dislikes, opinions
  • goals: Objectives, aspirations
  • projects: Active projects, status
  • technical: Tech stack, tools, dev environment
  • decisions: Important decisions, reasoning
  • infrastructure: Servers, IPs, services, ports, network config
  • assistant: Behavior rules, persona, communication style
  • relationships: People mentioned, their roles
  • routines: Habits, schedules
  • life_events: Milestones, changes
  • lessons: Insights, mistakes
  • work: Job context, career
  • health: Wellness info

Best Practices

  1. Auto-capture handles most storage — don't manually store what's already in conversation
  2. Use memory_store_raw for exact data: IP addresses, commands, configs
  3. Use category filter in memory_search to narrow results
  4. "Remember" keyword — when user says "remember X", the extraction prompt prioritizes it
  5. Sleep mode processes old conversations nightly — no need to re-extract from today's chat

CLI Commands

bash
openclaw mem0 search "query"          # Search memories
openclaw mem0 stats                   # Memory statistics
openclaw mem0-sleep                   # Run sleep maintenance manually
openclaw mem0-sleep --dry-run         # Preview what would be processed