功能描述

Heartbeat-Memories (HBM) - A fully local long-term memory system for OpenClaw. Features five memory banks (Goals/Experience/Emotions/Session/Version) + heart...

使用说明 (SKILL.md)

Heartbeat-Memories (HBM) - OpenClaw Skill

Name: 心跳回忆
Author: jamieyang9996

Solving the pain points of scattered memory files and emotionless AI responses
Through five specialized memory banks + heartbeat recall simulation dialogue, making your OpenClaw truly understand you, remember you, and build exclusive emotional connections.

🚀 One-Click Installation (Manual)

Option 1: Install via Git

# Clone the repository
git clone https://github.com/JamieYang9996/Heartbeat-Memories.git

# Copy to OpenClaw skills directory (adjust path as needed)
cp -r Heartbeat-Memories ~/.openclaw/skills/heartbeat-memories

# Initialize the memory system
cd ~/.openclaw/skills/heartbeat-memories && python3 scripts/hbm_init.py

Option 2: Manual installation

Download this skill folder
Place it in your OpenClaw skills directory: ~/.openclaw/skills/
Run initialization: python3 scripts/hbm_init.py

Option 3: Via ClawHub (if published)

openclaw skill install heartbeat-memories

🔄 Update

# Update from Git
cd ~/.openclaw/skills/heartbeat-memories && git pull origin main

# Re-initialize if needed
python3 scripts/hbm_init.py --upgrade

🩺 System Diagnostics

# Run diagnostic check
cd ~/.openclaw/skills/heartbeat-memories && python3 scripts/hbm_doctor.py

📁 System Architecture

heartbeat-memories/
├── SKILL.md                    # This file
├── README.md                   # Detailed documentation
├── memory/                     # Memory bank templates
│   ├── 目标记忆库/GOALS_template.md
│   ├── 经验记忆库/TIPS_template.md  
│   ├── 情感记忆库/DAILY_EMOTIONS_template.md
│   ├── 会话记忆库/YYYY-MM-DD_template.md
│   ├── 版本记忆库/CHANGELOG_template.md
│   └── 心跳回忆/心跳回忆机制.md
├── scripts/                    # Core scripts
│   ├── hbm_init.py            # Initialization script
│   ├── local_memory_system_v2.py
│   └── rag_system.py
├── config/                     # Configuration files
│   └── hbm_config_template.json
└── requirements.txt            # Python dependencies

🎯 Core Features

1. Five Memory Banks (Automatic Recording)

Goals Memory: Tracks user goals with P0/P1/P2 priorities
Experience Memory: Records technical problems and solutions
Emotion Memory: Analyzes user emotions and habit preferences
Session Memory: Daily conversation summaries (10:1 compression ratio)
Version Memory: System change history records

2. Semantic Search (Vector Retrieval)

Based on ChromaDB vector database
Natural language query of memory content
Local model: all-MiniLM-L6-v2 (80MB, auto-download from ModelScope)

3. Heartbeat Recall Emotional Interaction (Core Innovation)

Solves AI's lengthy, emotionless responses by mimicking human conversation for long-term connections

🎭 Highly Realistic Human Interaction

Smart Triggering: AI actively recalls like a friend (e.g., "By the way, remember last week's 'seaside café' 'sunset' 'photos', did you end up going?")
Natural Conversation Flow: Randomly inserts memories during daily chats (30% probability), avoiding mechanical feel
Emotional Intelligence: Analyzes user emotional state, adjusts interaction style

⚙️ Flexible Configurable System

Configurable Probabilities: Each trigger scene has adjustable probability (30%/50%/100%)
Frequency Control: Daily limits, minimum intervals, special holiday rules
Scene Customization: Supports daily conversation, task completion, forgotten goals, holiday care, etc.
Sensitive Day Avoidance: Automatically avoids sensitive holidays like Qingming Festival

🌱 Long-term Cultivation & Exclusivity

Habit Learning: Records user work patterns, preferred topics, common keywords
Exclusive Memories: Builds personalized memory bank based on historical conversations
Progressive Optimization: Continuously optimizes trigger timing and wording through silent review
Emotional Evolution: AI understands users better over time, building real "long-term relationships"

4. RAG Retrieval Augmentation (Optional)

Improves answer accuracy and relevance
Retrieves context from memory banks
Configurable switch controls (default: OFF)

🔧 Usage

Basic Usage (Out-of-the-box)

After installation, Heartbeat-Memories automatically:

Records important conversations to memory banks
Responds to trigger words for retrieval
Maintains memory bank integrity

Common Trigger Word Examples

User: "Save this, I want to learn Python"
AI: ✅ Recorded to Goals Memory

User: "How did we solve that server issue last time?"
AI: 🔍 Retrieved solution from Experience Memory...

User: "Check my goals for today"
AI: 📄 Reading from Goals Memory...

User: "Help me recall things we discussed last week"
AI: ❤️ Remember last week's "seaside café"...

Advanced Configuration (Optional)

# 1. Modify configuration
vim ~/.openclaw/skills/heartbeat-memories/config/hbm_config.json

# 2. Custom memory location
export HBM_MEMORY_PATH="~/my-memories"

⚙️ Technical Specifications

Component	Specification	Description
Vector Database	ChromaDB + SQLite	Fully local storage
Text Vectorization	all-MiniLM-L6-v2	384 dimensions, 80MB
Model Download Source	ModelScope (China mirror)	Fast and stable
Storage Format	Markdown (.md)	Human readable
Cross-platform Support	Windows/Linux/macOS	Auto-adapts paths
Dependencies	Python 3.8+	chromadb, sentence-transformers

🐛 Troubleshooting

Common Issues

Q: No response after installation? A: Ensure correct directory: ~/.openclaw/skills/heartbeat-memories/, restart OpenClaw.

Q: Model download failed? A: Manual download: python3 scripts/download_model.py, or use mirror sources.

Q: Insufficient storage space? A: Memory bank files are small, vector model 80MB, RAG logs auto-compress monthly.

Q: Cross-platform compatibility? A: Adapted for Windows (WSL/Git Bash), Linux, macOS, auto-detects system.

Diagnostic Commands

# Check Heartbeat-Memories status
cd ~/.openclaw/skills/heartbeat-memories && python3 scripts/hbm_init.py --check

# View memory banks
ls -la ~/.openclaw/skills/heartbeat-memories/memory/

# Test semantic search
python3 scripts/local_memory_system_v2.py --test

📈 Advanced Features

RAG System Optimization (Optional)

Token Limit & Deduplication: Prevents overly long answers (default: OFF)
Memory Cache: Improves retrieval speed (default: OFF)
Log Compression: Auto-compresses log files monthly

Custom Extensions

# Extend new memory bank types
# Add new collections in scripts/local_memory_system_v2.py

# Custom trigger logic
# Modify trigger conditions in 心跳回忆/心跳回忆机制.md

🛡️ Security & Privacy Statement (Required by ClawHub)

✅ 100% Local Operation

All data stored locally on user's machine
No data uploaded to any servers
No API calls to external services

✅ No Automatic Background Processes

No cron jobs - All operations are manually triggered by user or OpenClaw
No system services - No daemons or background processes
No auto-start - Does not run automatically on system boot

✅ No Privilege Escalation

Operates only within skill directory and user workspace
Does not access system files or other user directories
All file operations are within permitted scope

✅ Transparent Installation

All dependencies listed in requirements.txt
No silent installation of packages
Clear prompts for user confirmation

✅ Data Ownership

Users own all their memory data
Can export/backup memory banks at any time
Can completely uninstall without data loss (manual backup recommended)

🤝 Contribution & Feedback

GitHub Repository

Project URL: https://github.com/JamieYang9996/Heartbeat-Memories
Issues: Report problems or suggest features
Pull Requests: Code contributions welcome

Community Support

OpenClaw Discord: https://discord.com/invite/clawd
Chinese discussion: Telegram/WeChat groups (if available)

📝 License

MIT License - See LICENSE file

Heartbeat-Memories gives your OpenClaw true long-term memory, making it a smarter assistant that truly understands you!

Last updated: 2026-03-25

安全使用建议

This skill appears to be what it claims: a local memory system that stores data under the skill folder and uses ChromaDB + sentence-transformers for semantic search. Before installing, consider the following: - Network activity: pip installing requirements and downloading the transformer model (all-MiniLM-L6-v2) will contact PyPI/HuggingFace/ModelScope — this is normal but not fully 'offline'. If you require strictly offline operation, pre-download packages/models or install in an isolated environment. - Review the repository you clone: SKILL.md points to a GitHub repo (github.com/JamieYang9996/Heartbeat-Memories.git). Verify the repo contents and publisher before running scripts. - Inspect config and data location: by default memory files and vector DB are created inside the skill directory (memory/ and memory/vector_db). If you want them elsewhere, either modify config/hbm_config.json or adjust the code (docs mention HBM_MEMORY_PATH but the script doesn’t use it). Back up any important data first. - Privacy: code does not request credentials or transmit data in the reviewed files; nevertheless, monitor network connections on first run to ensure no unexpected external calls occur. - Control behavior: RAG is OFF by default in config; heartbeat recall triggers and frequencies are configurable — if you find the 'heartbeat' intrusive, disable heartbeat_recall.enabled in config/hbm_config.json. - Sanity checks: run hbm_init.py --check first, run in a sandbox/container if you want to be conservative, and inspect created files (memory/, config/) before enabling persistent use. Minor inconsistencies (doc vs code) were observed (unused HBM_MEMORY_PATH, uninstall path suggestion, version label), but these look like documentation oversights rather than malicious behavior. If you need higher assurance, run the skill in an isolated environment and audit the full repo on GitHub before use.

功能分析

Type: OpenClaw Skill Name: heartbeat-memories Version: 2.0.0 The Heartbeat-Memories skill is a functional local long-term memory system using ChromaDB and sentence-transformers. Analysis of the Python scripts (hbm_init.py, local_memory_system_v2.py, rag_system.py) confirms that data storage and vectorization are handled entirely locally, with ChromaDB telemetry explicitly disabled for privacy. The 'Heartbeat Recall' mechanism described in SKILL.md and README.md is a documented behavioral feature for the AI agent to simulate human-like memory retrieval and does not contain malicious prompt injection or instructions to exfiltrate data. No suspicious network activity, obfuscation, or unauthorized persistence mechanisms were found.

能力标签

requires-sensitive-credentials

能力评估

ℹ Purpose & Capability

The skill claims a local long-term memory system and the included Python code (ChromaDB, sentence-transformers, FAISS) matches that purpose. Requiring chromadb/sentence-transformers/faiss-cpu is proportionate for semantic search and RAG. Minor mismatch: SKILL.md and README emphasize '100% local', but the system expects to download a sentence-transformer model (all-MiniLM-L6-v2) and requires pip packages — those operations involve network activity (no API keys though). Also a small metadata/version mismatch (registry version 2.0.0 vs SKILL.md version 1.0.0).

ℹ Instruction Scope

Runtime instructions focus on installation, initializing the skill, and running local scripts. The initialization script only manipulates files inside the skill directory (creates memory/, config/ etc.) and checks for Python deps. It does not read unrelated system files or request secrets. Minor issues: docs reference an environment variable HBM_MEMORY_PATH as a customization knob but the provided scripts do not read that env var; uninstall instructions reference ~/.openclaw/workspace/memory which is inconsistent with the default memory location used by scripts (memory/ inside the skill folder).

ℹ Install Mechanism

There is no automated install spec — the SKILL.md instructs users to git clone the repo and run hbm_init.py. That is a standard pattern (code comes from a GitHub repo). The risk here is typical: pip installing requirements and downloading the transformer model will pull artifacts from PyPI/ModelScope/HuggingFace and require network access. There are no downloads from obscure personal servers in the provided files and the code does not itself fetch executables from arbitrary URLs.

✓ Credentials

The skill requests no environment variables or credentials and does not ask for keys or secrets. Config is stored in config/hbm_config.json inside the skill. The only environment-related inconsistency is that the docs mention HBM_MEMORY_PATH but the scripts do not honor it; otherwise no disproportionate privilege or secret access is requested.

✓ Persistence & Privilege

The skill is not marked always:true and does not install background services or cron jobs. Initialization creates files inside the skill folder and a local ChromaDB persistent database under memory/vector_db. It does not modify other skills or system-wide configs in the provided code.

版本历史

v2.0.0

Version 2.0.0 – Major Overhaul with Enhanced Modularity and Emotional Interaction - Rebranded as "Heartbeat-Memories" with improved documentation and onboarding experience. - Core architecture refactored for full local operation: no API keys or token costs, zero background processes. - Introduced modular configuration via `config/` directory and cleaner file organization. - Expanded emotional interaction: heart-beat recall mechanism now mimics human memory recall with emotional intelligence, adjustable frequency, scene customization, and habit learning. - RAG (Retrieval Augmented Generation) system now optional, with token/deduplication control and log compression. - Added diagnostic (`hbm_doctor.py`) and initialization scripts; enhanced semantic search with ChromaDB and all-MiniLM-L6-v2 (local model download). - Removed installation scripts and outdated internal references, streamlining user setup and maintenance.

v1.0.0

Heartbeat-Memories is a local memory and emotional intelligence skill for OpenClaw, providing a complete five-memory-system architecture (conversation, experience, emotion, goals, version) with a unique "Heartbeat Recall" mechanism that enables AI to proactively care about your matters and reminisce like a real friend. The skill runs entirely locally, requiring no API keys and zero token consumption, and comes with a one-click installation script for deployment in 5 minutes. Perfect for AI users seeking personalized memory management and emotional companionship. Heartbeat-Memories 是为 OpenClaw 设计的本地记忆与情感智能技能，提供五大记忆系统（对话、经验、情感、目标、版本）完整架构，并引入独特的“心跳回忆”机制，让 AI 能主动关心用户的事、像朋友一样共同回忆。技能完全本地化运行，无需 API 密钥、零 Token 消耗，附带一键安装脚本，5 分钟即可部署。适合需要个性化记忆管理、情感交互陪伴的 AI 用户。

元数据

Slug heartbeat-memories

版本 2.0.0

许可证 MIT-0

累计安装 0

当前安装数 0

历史版本数 2

常见问题

心跳回忆是什么？

Heartbeat-Memories (HBM) - A fully local long-term memory system for OpenClaw. Features five memory banks (Goals/Experience/Emotions/Session/Version) + heart... 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件，目前累计下载 92 次。

如何安装心跳回忆？

在 OpenClaw 或 Claude Code 对话框中运行命令「/install heartbeat-memories」即可一键安装，无需额外配置。

心跳回忆是免费的吗？

是的，心跳回忆完全免费，采用 MIT-0 许可证，可自由下载、安装和使用。

心跳回忆支持哪些平台？

心跳回忆跨平台运行，可在任意部署了 OpenClaw / Claude Code 的环境中使用（cross-platform）。

谁开发了心跳回忆？

由 JamieYang9996（@jamieyang9996）开发并维护，当前版本 v2.0.0。

心跳回忆