## Introduction
Zhipu AI has launched AutoClaw (澳龙), a self-evolving AI agent that introduces a groundbreaking memory system. Unlike traditional AI assistants that reset between conversations, AutoClaw learns from interactions and continuously improves. This review explores how this self-evolution mechanism works and why it matters for AI productivity.
## What is AutoClaw?
AutoClaw is Zhipu AI’s latest AI agent featuring:
– **Self-evolution mechanism**: Learns and improves from user feedback
– **Persistent memory**: Remembers corrections, preferences, and lessons
– **Skill Store**: Marketplace for specialized AI skills
– **Multi-platform support**: Works across various messaging platforms
## Key Features
### 1. Self-Evolution Mechanism
The core innovation addresses AI’s “forgetfulness” problem:
**Automatic Learning**
– After each conversation, AutoClaw scans for corrections
– Identifies user preferences and working styles
– Detects “踩坑经验” (lessons from mistakes)
– Displays valuable learnings as “Evolution Request” cards
**User-Controlled**
– All evolutions require user approval
– Users can ask “What have you learned recently?”
– Weekly evolution limit (1-3) ensures quality over quantity
– No noisy memorization (e.g., won’t remember “you like hotpot”)
**Trigger Methods**
– **Intent keywords**: “以后” (from now on), “记住” (remember), “永远” (always)
– **Complex task completion**: Learns from successful multi-step workflows
### 2. Skill Store: GLM Office Skills
The Skill Store launches with five professional tools:
**GLM Office Skills (5 Tools)**
– **PPT**: Professional presentation generation
– **DOCX**: Document creation with formatting
– **XLSX**: Spreadsheet and data visualization
– **PDF**: PDF conversion and editing
– **Charts**: Data visualization and infographics
**Technical Improvements**
– Built on GLM-5.1 with office-specific optimizations
– Scenario-specific design (academic papers, contracts, resumes)
– Smart self-check before delivery
– Format conversion (markdown↔doc↔pdf↔excel)
– Enhanced design system for more polished output
**Example Command:**
> “Help me create a competitive analysis PPT for new energy vehicles, plus speaker notes”
AutoClaw will research, then simultaneously generate a chart-rich PPT and corresponding DOCX speaker notes.
### 3. Expert Co-Created Skills
First batch includes community-contributed skills:
– **女娲Skill**: Creative writing assistance
– **横纵分析法**: Horizontal/vertical analysis framework
– **FrontendSlides**: Frontend development presentation tool
## Pricing
AutoClaw’s pricing model:
| Plan | Price | Features |
|——|——-|———-|
| Free | $0 | Basic self-evolution, limited skills |
| Pro | ~$10/month | Full Skill Store, priority evolution |
| Enterprise | Custom | API access, custom skills, support |
**GLM Coding Plan** (for developers):
– Starting at $3/month
– API access for integration
## Pros and Cons
### Pros
– **True memory**: No need to repeat context between sessions
– **Continuous improvement**: Gets better the more you use it
– **User control**: All evolutions require approval
– **Quality-focused**: Limits noisy memorization
– **Professional skills**: Office tools built on strong GLM foundation
### Cons
– **Learning curve**: Requires understanding evolution triggers
– **Privacy considerations**: Long-term memory of user preferences
– **Limited availability**: Primarily Chinese market initially
– **Quality variance**: Evolutions depend on usage patterns
## Comparison with Alternatives
| Feature | AutoClaw | ChatGPT | Claude |
|———|———-|———|——–|
| Self-Evolution | ✅ Yes | ❌ No | ❌ No |
| Persistent Memory | ✅ Yes | ⚠️ Limited | ⚠️ Session-only |
| Automatic Learning | ✅ Yes | ❌ No | ❌ No |
| Office Skills | ✅ Built-in | ⚠️ Plugins | ⚠️ Limited |
| Open Source | ⚠️ Partial | ❌ No | ❌ No |
## Who Should Use It?
**Ideal for:**
– Long-term project work requiring consistent context
– Professionals working with recurring tasks
– Users who frequently correct AI outputs
– Teams needing customized AI behavior
– Anyone tired of repeating instructions
## How It Differs from Traditional AI
**Traditional AI:**
“`
User: Write a professional email
AI: [Generates email]
User: Make it shorter
AI: [Generates again]
User: Still too formal, add a joke
AI: [Generates again]
// Next conversation: AI forgets everything
“`
**AutoClaw:**
“`
User: Write a professional email
AI: [Generates email]
User: Make it shorter
AI: [Generates again]
User: Still too formal, remember I like casual tone
AI: [Evolution request: “Remember casual tone preference”]
User: Approved
// Next conversation: AI automatically uses casual tone
“`
## Conclusion
AutoClaw represents a fundamental shift in AI interaction — from stateless responses to persistent, evolving assistance. The self-evolution mechanism solves a real pain point: the need to repeatedly explain preferences and corrections.
The Skill Store, powered by GLM-5.1’s strong foundation, adds practical productivity tools that make AutoClaw immediately useful. For users who want an AI that truly learns and improves, AutoClaw offers something unique in the market.
**Rating: 4.5/5**
—
*Want an AI that gets smarter the more you use it? AutoClaw’s self-evolution mechanism might be exactly what you’ve been waiting for.*
**Focus Keywords**: AutoClaw review, self-evolving AI, AI agent with memory, Zhipu AI AutoClaw