A powerful command-line interface for interacting with DeepSeek's AI models.

@PierrunoYT/deepseek-cli

• 🤖 Multiple Model Support

  • DeepSeek-V3 (deepseek-chat)
  • DeepSeek-R1 (deepseek-reasoner)
  • DeepSeek Coder (deepseek-coder)

• 🔄 Advanced Conversation Features

  • Multi-round conversations with context preservation
  • System message customization
  • Conversation history tracking
  • Context caching for better performance
  • Inline mode for quick queries

• 🧪 Beta Features

  • Prefix Completion: Complete assistant messages from a given prefix
  • Fill-in-the-Middle (FIM): Complete content between a prefix and suffix
  • Context Caching: Automatic caching for better performance

• 🛠️ Advanced Controls

  • Temperature control with presets
  • JSON output mode
  • Streaming responses
  • Function calling
  • Stop sequences
  • Top-p sampling
  • Frequency and presence penalties

• 📦 Package Management

  • Automatic version checking
  • Update notifications
  • Easy installation and updates
  • Development mode support

You can install DeepSeek CLI in two ways:

pip install deepseek-cli

git clone https://github.com/PierrunoYT/deepseek-cli.git
cd deepseek-cli
pip install -e .

To update to the latest version:

pip install --upgrade deepseek-cli

For development installation, pull the latest changes and reinstall:

git pull
pip install -e . --upgrade

The CLI will automatically check for updates on startup and notify you when a new version is available.

Set your DeepSeek API key as an environment variable:

export DEEPSEEK_API_KEY="your-api-key"

set DEEPSEEK_API_KEY="your-api-key"

To make it permanent, add it to your environment variables through System Settings.

DeepSeek CLI supports two modes of operation: interactive mode and inline mode.

After installation, you can start the CLI in interactive mode in two ways:

deepseek

deepseek
# or
python -m deepseek_cli

You can also use DeepSeek CLI in inline mode to get quick answers without starting an interactive session:

# Basic usage
deepseek -q "What is the capital of France?"

# Specify a model
deepseek -q "Write a Python function to calculate factorial" -m deepseek-coder

# Get raw output without token usage information
deepseek -q "Write a Python function to calculate factorial" -r

# Combine options
deepseek -q "Write a Python function to calculate factorial" -m deepseek-coder -r

Available inline mode options:

• -q, --query: The query to send to the model
• -m, --model: The model to use (deepseek-chat, deepseek-coder, deepseek-reasoner)
• -r, --raw: Output raw response without token usage information

• If the API key is not recognized:

  • Make sure you've set the DEEPSEEK_API_KEY environment variable
  • Try closing and reopening your terminal
  • Check if the key is correct with: echo $DEEPSEEK_API_KEY (Unix) or echo %DEEPSEEK_API_KEY% (Windows)

• If you get import errors:

  • Ensure you've installed the package: pip list | grep deepseek-cli
  • Try reinstalling: pip install --force-reinstall deepseek-cli

• For development installation issues:

  • Make sure you're in the correct directory
  • Try: pip install -e . --upgrade

Basic Commands:

• /help - Show help message
• /models - List available models
• /model X - Switch model (deepseek-chat, deepseek-coder, deepseek-reasoner)
• /clear - Clear conversation history
• /history - Display conversation history
• /about - Show API information
• /balance - Check account balance

Model Settings:

• /temp X - Set temperature (0-2) or use preset (coding/data/chat/translation/creative)
• /freq X - Set frequency penalty (-2 to 2)
• /pres X - Set presence penalty (-2 to 2)
• /top_p X - Set top_p sampling (0 to 1)

Beta Features:

• /beta - Toggle beta features
• /prefix - Toggle prefix completion mode
• /fim - Toggle Fill-in-the-Middle completion
• /cache - Toggle context caching

Output Control:

• /json - Toggle JSON output mode
• /stream - Toggle streaming mode
• /stop X - Add stop sequence
• /clearstop - Clear stop sequences

Function Calling:

• /function {} - Add function definition (JSON format)
• /clearfuncs - Clear registered functions

• 64K context length (64,000 tokens)
• Default max output: 4096 tokens
• Beta max output: 8192 tokens (requires beta mode)
• Supports all features
• General-purpose chat model
• Latest improvements:
  • Enhanced instruction following (77.6% IFEval accuracy)
  • Improved JSON output (97% parsing rate)
  • Advanced reasoning capabilities
  • Role-playing capabilities

• 64K context length
• 8K output tokens
• 32K Chain of Thought output
• Excels at complex reasoning
• Unsupported features: function calling, JSON output, FIM
• Unsupported parameters: temperature, top_p, presence/frequency penalties

• Default max output: 4096 tokens
• Beta max output: 8192 tokens (requires beta mode)
• Optimized for code generation
• Supports all features

Use XML-style tags to define the gap:

<fim_prefix>def calculate_sum(a, b):</fim_prefix><fim_suffix>    return result</fim_suffix>

Forces model to output valid JSON. Example system message:

{
    "response": "structured output",
    "data": {
        "field1": "value1",
        "field2": "value2"
    }
}

• Automatically caches context for better performance
• Minimum cache size: 64 tokens
• Cache hits reduce token costs
• Enabled by default

• coding: 0.0 (deterministic)
• data: 1.0 (balanced)
• chat: 1.3 (creative)
• translation: 1.3 (creative)
• creative: 1.5 (very creative)

• Automatic retry with exponential backoff
• Rate limit handling
• Clear error messages
• API status feedback

For support, please open an issue on the GitHub repository.

This project is licensed under the MIT License - see the LICENSE file for details.