什么是 hexmem？

hexmem 是一个强大的、由 SQLite 驱动的内存基座，旨在超越简单的聊天记录，走向真正的智能体身份和持久知识。通过为事实、事件和教训提供结构化数据库，它允许智能体保持一致的自我意识，并在多个会话中检索关键信息。它实现了一个模拟人类认知过程的分层内存模型，包括基于重要性的显著性和自然衰减。

作为复杂工作流的核心组件，hexmem 无缝集成到智能体环境中。它通过主语-谓语-宾语三元组支持复杂的数据关系，并提供语义搜索功能，确保您的 Openclaw Skills 可以在需要时准确访问相关上下文。

下载入口:https://github.com/openclaw/skills/tree/main/skills/santyr/skill-hexmem

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

hexmem 应用场景

• 在断开连接的会话中保持智能体的身份和核心价值观。
• 记录重大项目决策和架构事件以备将来参考。
• 构建智能体与之交互的实体、系统和人员的知识图谱。
• 跟踪长期目标和任务进度而不会丢失上下文。
• 捕捉汲取的教训，以避免在编码或部署中重复过去的错误。

1. 智能体使用迁移脚本初始化 SQLite 数据库，为身份、事实和事件建立结构化模式。
2. 定义关键身份属性和核心价值观，以播种智能体的角色和论理框架。
3. 通过提供的 bash 助手和 CLI 工具，将实时交互捕获为事件、事实或任务。
4. 衰减算法优先处理频繁访问或具有情感意义的数据，确保重要记忆保持活跃，而旧事实被取代。
5. 语义搜索和 SQL 视图允许智能体在每次会话开始时综合当前状态和历史，以实现完全的连续性。

hexmem 配置指南

使用以下命令直接将 hexmem 安装为您的 Openclaw Skills 之一：

clawhub skill install hexmem

或者，克隆存储库并手动初始化数据库：

git clone https://github.com/hexdaemon/hexmem.git
cd hexmem
./migrate.sh up

要在会话中启用辅助函数，请 source Shell 脚本：

source ~/your-workspace/hexmem/hexmem.sh

hexmem 数据架构与分类体系

hexmem 将数据组织到专门的表中，以确保高保真检索和长期存储：

name: hexmem
description: Structured memory database for AI agent identity, tasks, events, lessons, and persistent continuity. Use whenever the user says remember this, log this, track this, what should I do next, recap, summarize, what changed, or when you need to store/retrieve decisions, incidents, fleet ops notes, reminders, TODOs, goals, values, or facts about people/systems/projects across sessions.

HexMem - Structured Memory Substrate

HexMem is a SQLite-based persistent memory system for agent identity, knowledge, and becoming. Not just logs—structured self-knowledge.

Installation

Clone to your workspace:

cd ~/your-workspace  # e.g., ~/clawd, ~/workspace, etc.
git clone https://github.com/hexdaemon/hexmem.git
cd hexmem
./migrate.sh up  # Initialize database

Or install as a skill from ClawHub:

clawhub skill install hexmem

Quick Start

Source the helpers at session start:

# Set database location (optional, defaults to ~/clawd/hexmem/hexmem.db)
export HEXMEM_DB="$HOME/your-workspace/hexmem/hexmem.db"

# Load helpers
source ~/your-workspace/hexmem/hexmem.sh

For convenience, add to your session startup (AGENTS.md or equivalent).

Core Patterns

1. Identity & Self-Knowledge

Store who you are, not just what you did:

# Set identity attributes
hexmem_identity_set "name" "YourName"
hexmem_identity_set "did" "did:cid:bagaai..."

# Add self-schemas (domain-specific self-beliefs)
hexmem_schema "coding" "python-expert" "I specialize in Python development" 0.8

# View current self-image
hexmem_self_image
hexmem_identity_summary

2. Facts About Entities

Store knowledge as subject-predicate-object triples:

# Add entity first
hexmem_entity "person" "Alice" "Project collaborator"

# Store facts
hexmem_fact "Alice" "timezone" "America/Denver"
hexmem_fact "ProductionServer" "capacity" "16GB"

# Facts with emotional weight (affects retention)
hexmem_fact_emote "ProjectGoal" "milestone" "first deployment" 0.8 0.7

# Query facts
hexmem_facts_about "Alice"
hexmem_fact_history "ProjectGoal"  # See how facts evolved

3. Memory Decay & Supersession

Facts decay over time unless accessed. Recent/frequent access keeps them hot:

# Access a fact (bumps to hot tier, resets decay)
hexmem_access_fact 42

# Replace a fact (preserves history)
hexmem_supersede_fact 42 "new value" "reason for change"

# View by decay tier
hexmem_hot_facts      # ≤7 days since access
hexmem_warm_facts     # 8-30 days
hexmem_cold_facts     # 30+ days

# Get synthesis for an entity (hot + warm facts)
hexmem_synthesize_entity "Sat"

Decay logic:

• Frequently accessed facts resist decay
• Emotionally weighted facts decay slower
• Old facts are never deleted, just superseded
• Query v_fact_retrieval_priority for importance-ranked facts

4. Events & Timeline

Log what happened:

# Basic event
hexmem_event "decision" "fleet" "Changed fee policy" "Set min_fee_ppm to 25"

# Event with emotional tagging
hexmem_event_emote "milestone" "autonomy" "First zap received" 0.9 0.6

# Query events
hexmem_recent_events 10
hexmem_recent_events 5 "fleet"
hexmem_emotional_highlights  # High-salience memories

5. Lessons Learned

Capture wisdom from experience:

hexmem_lesson "lightning" "Channels need time to build reputation" "from fleet experience"
hexmem_lesson "debugging" "Check your own setup first" "Archon sync incident"

# Query lessons
hexmem_lessons_in "lightning"
hexmem_lesson_applied 7  # Mark lesson as used

6. Goals & Tasks

# Add goal
hexmem_goal "project-launch" "Ship v1.0 by Q2" "professional" 8
hexmem_goal_progress 1 25  # Update progress to 25%

# Add task
hexmem_task "Review pull requests" "Weekly review" 7 "2026-02-07"

# Check what needs attention
hexmem_pending_tasks

7. Semantic Search

Search memories by meaning, not just keywords:

hexmem_search "identity and autonomy"
hexmem_search "Lightning routing lessons" "lessons" 5

Setup required (one-time):

cd $HEXMEM_ROOT  # wherever you installed hexmem
source .venv/bin/activate
python embed.py --process-queue  # Generate embeddings for new content

Mandatory Defaults (Active Use)

When this skill is in play, behave as if HexMem is the source of truth for continuity.

Always do at the start of any ops / admin / debugging task

# Fast check (preferred)
/home/sat/clawd/hexmem/scripts/hexmem-check.sh

Or manually:

source ~/clawd/hexmem/hexmem.sh
hexmem_pending_tasks
hexmem_recent_events 5

Always do when the user says "remember" / "track" / "log"

Write it immediately as a task, fact, lesson, or event (don’t defer):

hexmem_event "note" "context" "
" "
"
# or
hexmem_task "