📖 Getting Started · Chapter 01

5 Core AI Concepts

Before using EasyClaw, spend 5 minutes understanding these concepts — they will help you truly understand how AI works, rather than blindly entering commands.

1. Agent (Intelligent Entity)

🗣️

Common Understanding

AI is your digital assistant or chatbot.

🧠

Deep Understanding

Regular AI answers questions directly; an Agent plans, uses tools, and verifies results before answering—like a truly capable colleague.

For example: when you say "help me organize my inbox", regular AI only tells you how to do it; EasyClaw's Agent directly opens the email client → classifies and archives → replies to important emails → reports completion status.

2. Skills (Capabilities)

🗣️

Common Understanding

AI capabilities are various advanced tools or apps.

🧩

Deep Understanding

Skills are packaged professional operation manuals. After installation, the Agent knows how to correctly call Lark API, generate images, recognize license plates... instead of guessing.

Think of a Skill as "instructions for AI": with it, AI masters the complete operation specifications of a certain domain. Skills market: clawhub.com

💡
How do Skills work?
When you send a task, the Agent scans descriptions of all installed Skills, loads the corresponding operation specifications after matching, and guides every subsequent step of execution.

3. Prompt (Instructions)

🗣️

Common Understanding

It's a "one-sentence request" you tell AI in natural language.

🎯

Deep Understanding

Prompt is the core interface for communicating with AI. A good Prompt includes: role definition, background information, specific step requirements, and output format.

TypeExampleEffect
❌ Normal Prompt"Help me write an email"AI can only guess, poor output quality
✅ Good Prompt"You are a B2B salesperson, help me write a product follow-up email to a CTO, professional and concise tone, ending with a clear next action"Role + background + format, precise output

4. Memory / MEMORY.md

🗣️

Common Understanding

AI's "notepad".

🗄️

Deep Understanding

Memory is the core of Agent's long-term memory. Regular conversations are cleared after ending; but content written to MEMORY.md will be prioritized by the Agent at every startup—it permanently remembers your preferences and work SOPs.

For example, if you tell the Agent "I prefer concise English responses, code in Python", after this record enters MEMORY.md, the Agent will automatically follow it every time without you needing to repeat.

Practical Advice
Tell your work habits, commonly used format requirements, important contacts, etc. to the Agent "once"—let it write to memory, and your AI assistant will understand you better over time.

5. Soul / SOUL.md

🗣️

Common Understanding

AI's "personality settings".

Deep Understanding

SOUL.md defines the Agent's behavior rules, values, and operational boundaries. It's the Agent's foundational constitution—what can be done, what absolutely cannot be done, all written here.

If Memory is "what to remember", then Soul is "rules of who to become". For example: only answer product-related questions, financial operations must be double-confirmed, prohibit requesting passwords...

⚠️
Advanced User Note
SOUL.md settings directly affect the Agent's safety boundaries. If you plan to deploy EasyClaw as a team tool, be sure to configure this file carefully.

Advanced Concepts (Optional)

The following three concepts are for advanced users who want to customize EasyClaw deeply. Beginners can skip and come back when needed.

🔀

Workflow (Process Automation)

Chain multiple Agent tasks in logical sequence to form a repeatable automation pipeline. Example: "Receive email → Extract key info → Generate summary → Send Lark notification" is a typical Workflow.

📦

JSON (Data Exchange Format)

Standard format for data transfer between Agent and external tools/APIs. When you ask Agent to call third-party services (like weather, calendar, database), data usually flows in JSON format. No manual writing needed, but understanding structure helps troubleshooting.

🔑

API Key (Access Credential)

Identity credential for accessing AI models or third-party services. EasyClaw uses official credit system by default, no need for own Key; advanced users can fill in their own API Key in "Add Custom Model" to access specified models.

💡
When do you need to care about these?
When you want to build complex multi-step automation processes (Workflow), access your own API Key to save costs, or need to troubleshoot data transfer issues between Agent and external tools (JSON), these three concepts become very useful.
🎉
Quick Concept Reference
Agent = AI colleague with execution power | Skills = Professional capability plugins | Prompt = High-quality communication method | Memory = Long-term memory repository | Soul = Behavioral constitution