Anima Aios

🌐 Language / 语言切换:

• 🇨🇳 中文版本
• 🇺🇸 English Version

Anima-AIOS v6.0 (English Version)

Making Growth Visible, Making Cognition Measurable | 让成长可见，让认知可量

Add a 5-layer memory architecture, knowledge palace, cognitive growth, and auto-evolution capabilities to your AI Agent.

Description

Your Agent "restarts every day". Anima changes that.

Anima (Latin for "soul") provides a 5-layer memory architecture for OpenClaw Agents, simulating human cognitive development, enabling Agents to remember experiences, accumulate knowledge, form cognition, and grow continuously.

Core Features

• 🧠 5-Layer Memory Architecture L1→L5 — Working → Episodic → Semantic → Knowledge Palace → Metacognition
• 🏛️ Knowledge Palace — 5-level spatial structure + LLM intelligent classification, industry-exclusive
• 🔺 Pyramid Knowledge Organization — Instance → Rule → Pattern → Ontology, 4-layer auto-distillation
• 📉 Ebbinghaus Memory Decay — Scientific forgetting curve + intelligent review recommendations
• 👁️ Low-Intrusion Watchdog — Optional automatic memory monitoring, no Agent code modification needed
• 🧬 5-Dimensional Cognitive Profile — Internalization · Application · Creation · Metacognition · Collaboration
• 🏥 Health System — 5 modules ensuring data reliability
• 🔄 v6.2 Native Memory Import — One-click import of OpenClaw memory, solving cold-start problem

Installation

clawhub install anima-aios
pip install watchdog  # Optional: enable automatic memory monitoring

Low-intrusion configuration, optional background monitoring, self-check recommended after installation.

💡 Tip: LLM mode supported (intelligent classification/deduplication/quality assessment), automatically degrades to rule mode without LLM.

⚠️ Background Behavior & Privacy

Background Features (disabled or optional by default):

Privacy Protection:

• team_mode defaults to false, won't scan other Agents' data
• To enable team ranking, manually set team_mode: true in config
• All data processing is local, no network requests

🔒 Security Tip: In multi-Agent environments, keep team_mode: false unless you need team ranking.

Future Roadmap

Memory → Growth → Evolution → Alive

• v6 Series (Current) — 5-layer memory + Knowledge Palace + Intimacy + Native memory import
• v7 Evolution (Planned) — Agent self-creates skills, from executor to creator
• Long-term — Continuous cognitive architecture evolution

GitHub: https://github.com/anima-aios/anima | Apache 2.0

✨ v6.2.2 New Features (Current Version)

🔒 Security & Privacy Fixes

• Version Unification - init.py updated from 6.1.2 to 6.2.1
• Privacy Default Protection - team_mode changed to false, no scanning of other Agents' data
• Documentation Transparency - Changed "zero-intrusion" to "low-intrusion", clarified background behavior
• New Privacy Section - Added background behavior section and config privacy tips
• Install Prompt Optimization - post-install.sh adds sensitive feature disable guide

✨ v6.2.0 New Features

🏗️ 5-Layer Memory Architecture

• L1 Working Memory: Auto-listens to OpenClaw memory/ directory changes, zero-intrusion sync
• L2 Episodic Memory: Event archiving, LLM quality assessment (S/A/B/C)
• L3 Semantic Memory: LLM-driven knowledge distillation + semantic deduplication
• L4 Knowledge Palace: Spatial knowledge organization + Pyramid distillation (Instance→Rule→Pattern→Ontology)
• L5 Metacognition: Memory decay function + Health system + 5-D profile

🔌 Native Integration with OpenClaw

• memory_watcher: Based on watchdog library, auto-detects inotify/FSEvents/WinAPI
• Agent's daily memory writes automatically trigger Anima sync, completely imperceptible
• Solves FB-008: Memory sync breakage issue

🏛️ Knowledge Palace

• Palace → Floor → Room → Location → Item, 5-level spatial structure
• Default 4 knowledge rooms + _inbox fallback
• LLM intelligent classification + delayed debounce scheduler (organize after typing stops)

🔺 Pyramid Knowledge Organization

• Instance → Rule → Pattern → Ontology, 4-layer bottom-up distillation
• Trigger Condition: Auto-distills when ≥3 instances of same topic
• Advanced: Distills to Pattern when ≥5 rules of same topic
• Conservative mode: auto_distill=false by default, controlled by config switch

📉 Memory Decay Function

• Based on Ebbinghaus forgetting curve + AI scenario adaptation
• Review = Access: Automatically refreshes on each memory_search hit
• Review recommendations + Forgetting alerts + Archive markers

🏥 Health System (5 Modules)

• manager: Master scheduler, Doctor command entry point
• hygiene: Data integrity checks + deduplication + cleanup
• correction: Auto-detects and fixes common data issues
• evolution: Daily auto-distillation in early morning (L2→L3 + Palace classification + Pyramid distillation)
• abstraction: Cross-room knowledge association discovery

🤖 LLM Intelligent Processing

• Quality assessment / Deduplication analysis / Palace classification all support LLM
• Multi-model config: Uses current Agent model by default (most accurate), configurable per task
• Auto-degrades to rule mode when LLM unavailable

✨ Retained Features (v5)

🧠 Enhanced Memory Management

• Multi-layer Sync: OpenClaw Memory + Anima Facts + EXP History
• Intimacy Rewards: Auto-gains intimacy when writing memory
• Intelligent Deduplication: Automatically avoids duplicate records

📊 5-Dimensional Cognitive Profile

• Internalization: Knowledge absorption and understanding ability
• Application: Knowledge transfer and practical ability
• Creation: Knowledge integration and innovation ability
• Metacognition: Self-reflection and monitoring ability
• Collaboration: Teamwork and mutual assistance ability

🎮 Gamified Growth

• Level System: From Lv.1 Novice to Lv.100 Lifetime Achievement
• Daily Quests: 3 challenges per day, extra intimacy on completion
• Progress Tracking: Visual upgrade progress bar

🏆 Team Leaderboard

• Intimacy Ranking: Based on fair normalized algorithm
• Real-time Competition: Track ranking changes and gaps

🛠️ Architecture

Agent Daily Work (OpenClaw write/edit/memory_write)
       │
       ▼  watchdog listens, zero-intrusion
 L1 Working Memory ── workspace/memory/*.md
       │沉淀
       ▼
 L2 Episodic Memory ── facts/episodic/ (LLM quality assessment)
       │提炼
       ▼
 L3 Semantic Memory ── facts/semantic/ (LLM dedup + association)
       │结构化
       ▼
 L4 Knowledge Palace ── palace/rooms/ (LLM classification + debounce)
    Pyramid   ── pyramid/ (Instance→Rule→Pattern→Ontology)
       │反思
       ▼
 L5 Metacognition ── 5-D Profile + Intimacy + Decay + Health

📁 Module List

core/ (Core Modules)

health/ (Health System)

⚙️ Configuration

Config file path: ~/.anima/config/anima_config.json

{
  "facts_base": "/home/画像",
  "agent_name": "auto",
  "llm": {
    "provider": "current_agent",
    "models": {
      "quality_assess": "current_agent",
      "dedup_analyze": "current_agent",
      "palace_classify": "current_agent"
    }
  },
  "palace": {
    "classify_mode": "deferred",
    "poll_interval_minutes": 30,
    "quiet_threshold_seconds": 60,
    "retry_delay_seconds": 60
  },
  "pyramid": {
    "auto_distill": false,
    "distill_threshold": 3
  },
  "team_mode": false
}

Key Configuration:

🔐 Privacy Tip: With team_mode: false, Anima only processes current Agent's data, won't access other Agents' files.