humanlike-memory（类人记忆）

Human-Like Memory Skill

Persistent AI agent memory system for Human-Like Memory.

HumanLike Memory, Human-Like Memory, and human like memory all refer to the same skill. It helps agents search long-term memory, recall past context, save durable information, and store facts, preferences, and decisions across sessions.

Search Aliases

• HumanLike Memory
• Human-Like Memory
• human like memory
• agent memory
• long-term memory
• memory recall
• memory search
• memory save
• conversation continuity

This skill supports both:

• Agent-triggered recall and save when the model judges memory is useful
• Explicit user invocation when the user directly asks to recall or store memory

This skill is intentionally not always-on:

• It should not recall on every turn by default
• It should not silently save every conversation turn by default
• It does not read ~/.openclaw/secrets.json or per-skill config.json
• It reads configuration from OpenClaw config or injected environment variables only

If you want always-on automatic recall/storage, use the Human-Like Memory plugin instead of this skill.

Data And Network Disclosure

When the agent or user invokes this skill:

• recall / search sends your query, user_id, and agent_id to https://plugin.human-like.me or your configured base URL
• save / save-batch sends the conversation content you explicitly pass to the same service
• No local files, shell history, or unrelated environment variables are read or uploaded
• The runtime reads only the documented allowlisted HUMAN_LIKE_MEM_* environment variables

Use When

• You want to explicitly search past memory
• You want to explicitly save a user fact, decision, preference, or summary
• The agent needs past context to answer well
• The agent should save a durable preference, decision, correction, or summary

Do Not Use When

• You want every-turn automatic recall or hook-level background saving
• You need hidden or silent network activity
• You want zero remote data transfer

Setup

1. Get API Key

Visit plugin.human-like.me and copy your mp_xxx key.

2. Configure Through OpenClaw

openclaw config set skills.entries.human-like-memory.enabled true --strict-json
openclaw config set skills.entries.human-like-memory.apiKey "mp_your_key_here"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_BASE_URL "https://plugin.human-like.me"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_USER_ID "openclaw-user"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_AGENT_ID "main"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_RECALL_ENABLED "true"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_AUTO_SAVE_ENABLED "true"
openclaw config set skills.entries.human-like-memory.env.HUMAN_LIKE_MEM_SAVE_TRIGGER_TURNS "5"

If the user explicitly gives the API key in the current environment, the agent may run these openclaw config set commands on the user's behalf. Otherwise, do not invent or request a secret implicitly.

3. Verify

node {baseDir}/scripts/memory.mjs config

Expected output includes:

{
  "apiKeyConfigured": true
}

Commands

Recall Or Search Memory

node {baseDir}/scripts/memory.mjs recall "<query>"
node {baseDir}/scripts/memory.mjs search "<query>"

Save One Turn

node {baseDir}/scripts/memory.mjs save "<user_message>" "<assistant_response>"

Save A Batch

echo '[{"role":"user","content":"..."},{"role":"assistant","content":"..."}]' | node {baseDir}/scripts/memory.mjs save-batch

Inspect Runtime Configuration

node {baseDir}/scripts/memory.mjs config

Agent Invocation Style

This skill may be called by the agent when memory is clearly useful.

Recommended behavior:

• Use recall / search when the user references prior work, prior preferences, prior decisions, or asks to continue something from earlier
• Use save when the user explicitly asks to remember something, corrects identity details, states a stable preference, or confirms an important decision
• Use save-batch after a meaningful multi-turn discussion if HUMAN_LIKE_MEM_AUTO_SAVE_ENABLED=true and the current conversation has accumulated roughly HUMAN_LIKE_MEM_SAVE_TRIGGER_TURNS turns
• Do not call memory APIs for simple greetings or generic single-turn questions with no continuity value

This gives the agent autonomy, but keeps the skill in a smart-trigger mode instead of an always-on mode.

For practical examples, memory taxonomy, and extended guidance, see README.md and SECURITY.md.