Thinking Capability

Feature Overview: Thinking Capability is iFlow CLI's intelligent reasoning system that triggers AI deep thinking through keywords, providing deeper analysis and reasoning.

Learning Time: 10-15 minutes

Prerequisites: iFlow CLI installed and configured, basic command line operations

What is Thinking Capability

Thinking Capability is iFlow CLI's built-in intelligent reasoning enhancement system that triggers AI model's deep thinking mode by recognizing specific keywords in user input. When activated, the AI will perform internal reasoning before giving final answers, showing its thought process to provide more accurate and deeper responses.

Core Features

Feature	Description	Advantage
Smart Keyword Recognition	Supports Chinese and English thinking trigger words	Natural language interaction
Multi-level Reasoning Intensity	5 reasoning levels: none, normal, hard, mega, ultra	Precise thinking depth control
Real-time Thinking Display	Shows AI's thinking process	Enhanced transparency and credibility
Multi-model Adaptation	Supports OpenAI o1, DeepSeek, GLM-4.5, etc.	Broad model compatibility
Flexible Display Modes	Full, compact, indicator display modes	Adapts to different usage scenarios

How It Works

Thinking Trigger Flow

User Input → Keyword Analysis → Intent Recognition → Config Generation → Model Call → Thinking Display
    ↓
[Contains thinking words] → [Regex matching] → [Reasoning level] → [Token limit] → [Deep reasoning] → [Process visualization]

Reasoning Level System

none: No thinking mode, direct response (0 tokens)
normal: Basic thinking, recognizes "想想", "think" and other basic words (2,000 tokens)
hard: Intermediate thinking, recognizes "再想想", "think harder" and other words (4,000 tokens)
mega: Advanced thinking, recognizes "好好思考", "think really hard" and other words (10,000 tokens)
ultra: Super thinking, recognizes "超级思考", "think super hard" and other words (32,000 tokens)

Detailed Feature Description

Keyword Trigger System

Chinese Trigger Words

Super Thinking (Ultra):

超级思考、极限思考、深度思考
全力思考、超强思考
认真仔细思考

Mega Thinking (Mega):

强力思考、大力思考、用力思考
努力思考、好好思考、仔细思考

Hard Thinking (Hard):

再想想、多想想
想清楚、想明白、考虑清楚

Basic Thinking (Normal):

想想、思考、考虑

English Trigger Words

Super Thinking (Ultra):

ultrathink
think really super hard
think intensely

Mega Thinking (Mega):

megathink
think really hard
think a lot

Hard Thinking (Hard):

think about it
think more
think harder

Basic Thinking (Normal):

think

Thinking Display

The thinking process is presented in the thinking status indicator.

✻ Thinking...

Model Support Information

Tab Switching Function Support

✅ glm-4.6: Fully supports Tab key switching between thinking/non-thinking modes
✅ deepseek-3.2: Fully supports Tab key switching between thinking/non-thinking modes
❌ Other models: Does not support Tab switching function, only supports keyword-triggered thinking mode

Feature Differences

Model	Tab Switching	Keyword Trigger	Thinking Mode Support
glm-4.6	✅ Supported	✅ Supported	✅ Full support
deepseek-3.2	✅ Supported	✅ Supported	✅ Full support
Other models	❌ Not supported	✅ Supported	⚠️ Limited support

Note: For models that do not support Tab switching, users need to input thinking keywords (such as "think", "想想", etc.) to trigger thinking mode.

Model Adaptation System

Thinking capability currently supports hybrid reasoning models and currently supports the glm-4.6 deepseek-3.2 models

Tab Switching Function

iFlow CLI provides the ability to quickly switch between thinking mode and non-thinking mode using the Tab key:

Supported Models

glm-4.6: Supports Tab switching between thinking/non-thinking modes
deepseek-3.2: Supports Tab switching between thinking/non-thinking modes
Other models: Does not support Tab switching function, only supports keyword-triggered thinking mode

Usage Method

During conversation input, press the Tab key to switch between thinking mode and non-thinking mode
When switching, the current mode status will be displayed above the input box
After switching, input content will be preserved and can continue to be edited

Important Notes

⚠️ Avoid frequently switching thinking modes within the same context session:

Different thinking modes use different internal processing logic
Frequent switching may lead to inconsistent context understanding
It is recommended to determine which mode to use when starting a new conversation and maintain consistency

Mode Status Indicators

Thinking Mode: Displays "Thinking Mode"
Non-Thinking Mode: Displays "Normal Mode"

Usage Examples

Basic Usage

# 
> Think deeply about this complex system design problem

English Usage

# Super thinking
> ultrathink this complex system design

Best Practices

Usage Recommendations

Context Switching Warning

⚠️ Important Safety Note: It is not recommended to frequently switch between thinking mode and non-thinking mode within the same conversation context.

Reasons:

Context Consistency: Thinking mode and non-thinking mode use different internal processing logic
Understanding Deviation: Frequent switching may cause the AI to have biased understanding of conversation history
Response Quality: May affect the coherence and accuracy of responses

Recommended Practices:

When starting a new conversation, determine whether to use thinking mode or non-thinking mode
Maintain mode consistency throughout the conversation process
If mode switching is needed, it is recommended to start a new conversation session
For complex tasks, it is recommended to use thinking mode throughout for deeper analysis

Troubleshooting

Thinking Not Triggered

Check if input contains correct trigger keywords
Verify if the model being used supports thinking capability
Confirm environment variable configuration is correct
Check if the current model supports Tab switching function (only glm-4.6 and deepseek-3.2 support)