CLI Reference

Command Line Usage

zhi                        # Interactive REPL
zhi -c "message"           # Single message mode
zhi run <skill> [files]    # Run a skill
zhi --setup                # Configuration wizard
zhi --version              # Show version
zhi --debug                # Enable debug logging
zhi --no-color             # Disable color output
zhi --language <LANG>      # Set interface language (auto, en, zh)

Modes

Interactive REPL

Launch the REPL by running zhi with no arguments:

zhi

Features:

• Persistent command history (saved to ~/.config/zhi/history.txt)
• Tab completion for slash commands, model names, and skill names
• Multi-line input with \ continuation
• CJK/IME input support via prompt_toolkit
• Sensitive inputs (containing api_key, password, token, secret) are excluded from history

Single Message Mode

Send a single message and exit:

zhi -c "What files are in the current directory?"

Skill Run Mode

Run a named skill with optional input files:

zhi run summarize report.txt
zhi run compare v1.md v2.md
zhi run extract-table invoices/receipt.pdf

Pipe Mode

Pipe text from stdin:

echo "translate this to Chinese" | zhi
cat document.txt | zhi -c "summarize this"
git log --oneline -20 | zhi -c "summarize recent changes"

Pipe detection

zhi automatically detects when stdin is not a terminal and reads from it.

---

Slash Commands

The REPL supports 14 slash commands for controlling the session:

Command	Description	Example
/help	Show available commands and tips	/help
/auto	Switch to auto mode (skip tool confirmations)	/auto
/approve	Switch to approve mode (confirm before risky tools)	/approve
/model <name>	Switch the LLM model for the session	/model glm-4-flash
/think	Enable thinking mode (extended reasoning)	/think
/fast	Disable thinking mode (faster responses)	/fast
/run <skill> [files]	Run a skill with optional file arguments	/run summarize report.txt
/skill list	List all available skills	/skill list
/skill show <name>	Show details of a specific skill	/skill show analyze
/status	Show current session state	/status
/reset	Clear conversation history (with confirmation)	/reset
/undo	Remove the last user message and AI response	/undo
/usage	Show token usage statistics	/usage
/verbose	Toggle verbose output	/verbose
/exit	Exit zhi (shows usage stats if any)	/exit

Permission Modes

Mode	Behavior	Switch command
approve (default)	Confirm before executing risky tools (file_write, shell, skill_create)	/approve
auto	Skip confirmation for risky tools (shell still always confirms)	/auto

Shell always confirms

The shell tool requires confirmation regardless of permission mode. This is a safety measure that cannot be bypassed.

Model Switching

Available models:

Model	Tier	Thinking	Tools
glm-5	Premium	Yes	Yes
glm-4-flash	Economy	No	Yes
glm-4-air	Economy	No	Yes

zhi> /model glm-4-flash
Model switched to glm-4-flash

Status Display

zhi> /status
Model: glm-5 | Mode: approve | Thinking: on | Verbose: off | Turns: 3 | Tokens: 1523

---

Global Options

Option	Description
--version	Print version and exit
--setup	Run the first-time configuration wizard
--debug	Set logging to DEBUG level (shows API calls, tool execution details)
--no-color	Disable Rich formatting and colors (also respects NO_COLOR env var)
--language LANG	Set interface language (auto, en, zh). Overrides config file setting.
-c MESSAGE	One-shot mode: send a single message, print the response, and exit

---

Exit

Exit the REPL using any of:

• /exit command
• Ctrl+D (EOF)
• Ctrl+C cancels current input but does not exit

On exit, zhi displays session token usage if any tokens were consumed.