Chat Usage¶

CLIver provides an interactive interface for communicating with various large language models. Chat is the default mode — no subcommand needed. This guide covers all the features and options available.

Basic Usage¶

To start an interactive session with the default model:

cliver

This will open an interactive session using your configured default LLM provider.

You can also pass a query directly:

cliver "What is the capital of China?"

Selecting Different Models¶

DeepSeek Models¶

To chat with DeepSeek models:

# Use DeepSeek-R1, note that you need to configure the model first
cliver --model deepseek-r1

QWen3 Models¶

To chat with QWen3 coder model:

# Use Qwen3-Coder, note that you need to configure the model first
cliver --model qwen3-coder

Chat Configuration Options¶

Temperature Control and others¶

Control the creativity of the model's responses:

Without query specified, it starts an interactive session with the options specified as the default value.

# More creative responses (higher temperature)
cliver --model deepseek-r1 --temperature 0.9

# More deterministic responses (lower temperature)
cliver --model deepseek-r1 --temperature 0.2

# Set max tokens for response
cliver --max-tokens 1024

# Set top_p parameter for sampling
cliver --top-p 0.9

# Set frequency penalty
cliver --frequency-penalty 0.5

System Prompt¶

Set a system prompt to guide the model's behavior:

NOTE: the system message will be appended to the builtin system message if specified.

cliver --system-message "You are a helpful assistant that responds in a professional manner."

Advanced Chat Features¶

Using MCP Servers¶

As long as MCP servers are configured, all tools will be included by default.

You can filter the tools using --included-tools option:

cliver --included-tools "*time"

Using Skills¶

CLIver has an LLM-driven skill system. During a chat session, the LLM can discover and activate skills automatically using the builtin skill tool, or you can activate them manually:

/skills run brainstorm design a login page

Skills are defined as SKILL.md files discovered from .cliver/skills/ (project), ~/.cliver/skills/ (global), and other compatible directories.

See Skills for details on creating and using skills.

File Integration¶

Work with files directly in the chat:

# Include a file in your message
cliver "Can you summarize this document?" --file /path/to/document.txt

# Process multiple files
cliver "Compare these two files" --file /path/to/file1.txt --file /path/to/file2.txt

Examples¶

Example 1: Professional Assistant Session¶

cliver \
  --model qwen3-coder \
  --system-message "You are a professional technical assistant. Provide concise, accurate answers with examples when possible." \
  --temperature 0.3

Example 2: Creative Writing Assistant¶

cliver \
  --model deepseek-r1 \
  --system-message "Help me brainstorm creative writing ideas. Be imaginative and provide detailed suggestions." \
  --temperature 0.8

Example 3: Code Review Session¶

cliver \
  --model qwen3-coder \
  --system-message "Review this code for best practices, security issues, and potential improvements." \
  --file /path/to/code.py

Next Steps¶

After mastering the chat command, learn about Skills for specialized task activation, Memory & Identity for agent personalization, or check out Workflows to automate complex multi-step operations.