CLI command reference

Use qwen --help in your shell and /help inside an interactive session. They are different interfaces. This reference covers root options visible in help, user slash commands, and extension subcommands. Hidden host protocols and developer diagnostics are not a stable public API.


qwen
qwen "Explain this project"
qwen -p "Summarize the README"
qwen --help
qwen --version

The positional argument is the initial prompt; -p selects print mode. Separate a prompt from variadic options with -- so it is not consumed as another option value. Examples require a configured model.

Root options

Syntax	Purpose
`-d, --debug`	Run in debug mode
`-m, --model <model>`	Model for the current session. Default and New Models use model name; Custom uses modelID
`--reasoning-effort <level>`	Set reasoning effort level
`--thinking <mode>`	Set thinking mode (auto, adaptive, enabled, or disabled)
`--thinking-budget <tokens>`	Set a fixed thinking token budget (implies —thinking enabled)
`--context-window <size>`	Explicit context window for supported models
`--list-models`	List available models for the current user
`-i, --prompt-interactive <text>`	Execute prompt and continue in interactive mode
`-w, --cwd <dir>`	Change working directory before startup
`--config-dir <dir>`	Use a custom user-level config root for this run
`--worktree [name]`	Start in a new git worktree
`--permission-mode <mode>`	Permission mode: default, accept_edits, bypass_permissions, dont_ask, auto (subject to policy).
`--dangerously-skip-permissions`	Skip ordinary approval requests; non-skippable checks and policies still apply
`--allowed-mcp-server-names <name>`	Allowed MCP server names
`--tools <tools...>`	Restrict available built-in tools ("" to disable all, “default” for all; use -p or — to separate from query)
`--allowed-tools <tool>`	Tools to allow
`--disallowed-tools <tool>`	Tools to deny
`--attachment <file>`	Attach files to the initial prompt
`--plugin-dir <dir>`	Plugin directories to load
`-c, --continue`	Continue the most recent session
`--fork-session`	Create a new session from a resumed conversation
`-r, --resume [id]`	Resume a previous session by identifier
`-n, --name <name>`	Set a display name for this session
`--session-id <id>`	Use a specific session ID
`--list-sessions`	List available sessions and exit
`--delete-session <index>`	Delete a session by index number
`--add-dir <dir>`	Additional directories to include in the workspace
`-o, --output-format <format>`	The format of the CLI output
`--input-format <format>`	The format of the CLI input
`--max-output-tokens <size>`	Set maximum model output tokens
`-p, --print`	Print response and exit (non-interactive)
`--no-session-persistence`	Disable session persistence - sessions will not be saved to disk and cannot be resumed (only works with —print)
`--agent <name>`	Agent for the current session
`--agents <json>`	JSON object defining custom agents
`--append-system-prompt <text>`	Append to the default system prompt
`--system-prompt <text>`	System prompt for the session
`--output-style <style>`	Output style for this session
`--max-model-request-retries <count>`	Maximum retries for Provider model requests
`--mcp-config <config>`	Load MCP servers from JSON file(s) or inline JSON
`--strict-mcp-config`	Only use MCP servers from —mcp-config
`--setting-sources <source>`	Setting sources to load (user, project, local)
`--settings <json>`	Load additional settings from a JSON file path or inline JSON string

The command framework also provides --help and --version. Tool allow rules do not replace project trust or organizational restrictions. Non-skippable checks still apply when bypassing approvals. Read permissions before using --dangerously-skip-permissions. Thinking and context-window options depend on model support.

Interactive slash commands

Command	Purpose
`/about`	Show version info
`/add-dir`	Add a directory to the workspace context
`/agents`	Manage agents
`/branch`	Create a new session branch from the current conversation
`/btw`	Ask a quick side question without interrupting the main conversation
`/clear`	Clear the screen and start a new conversation
`/commands`	Reload and list all available slash commands
`/compact`	Compresses the context by replacing it with a summary
`/context`	Visualize current context window usage
`/context-window`	Set context window for the current model
`/continue`	Continue the most recent session for the current project
`/copy`	Copy the last assistant response to clipboard, or /copy N for the Nth-latest
`/crontab`	Manage scheduled loops and cron jobs
`/diff`	Show uncommitted git changes
`/editor`	Set external editor preference
`/effort`	Set reasoning effort for the current model
`/export`	Export the current conversation to a file or clipboard
`/goal`	Set or manage a persistent goal for the current session
`/help`	Show application help and keyboard shortcuts
`/hooks`	Open hooks management panel
`/init`	Analyzes the project and creates a tailored context file
`/insights`	Generate a report analyzing your sessions
`/mcp`	Manage configured Model Context Protocol (MCP) servers
`/memory`	Commands for interacting with memory
`/model`	Set or manage model configuration
`/new`	Start a new conversation
`/output-style`	Configure the output style for the next conversation
`/peers`	List peer sessions and review held cross-session messages
`/permissions`	Manage permissions
`/plan`	Toggle Plan Mode (enter when off, exit when on)
`/plugins`	Manage plugins
`/profile`	Inspect or control runtime profiling (diagnostics)
`/quit`	Exit the application
`/release-notes`	Show recent release notes
`/rename`	Rename the current conversation
`/resume`	Resume a session by identifier, or open the session browser
`/review`	Review code changes and find actionable issues
`/rewind`	Open rewind selector
`/settings`	View and edit application settings
`/shortcuts`	Toggle the shortcuts panel above the input
`/skills`	Manage agent skills
`/status`	Show session status
`/statusline`	Set up a custom status line. Uses a sub-agent to read your shell config and create a statusline script that’s updated after each response.
`/subtask`	Send a subagent off with your full context; its result comes back here
`/tasks`	Show background tasks panel
`/theme`	Change the theme
`/tools`	List available tools
`/usage`	Show usage statistics for the current session
`/vim`	Toggle vim mode on/off for this session
`/workflows`	Browse workflow tasks in the session rail
`/auth`	Connect or switch an account/provider; aliases `/connect`, `/login`
`/logout`	Sign out of the selected account; alias `/signout`
`/terminal-setup`	Configure terminal integration in supported interactive terminals

Common aliases include /summarize and /compress for /compact, /checkpoint for /rewind, /bg and /background for /tasks, /config for /settings, and /exit for /quit.

This list does not imply every command is enabled in every environment: planning and agents can be disabled; /subtask requires an initial turn and does not allow a nested fork; /terminal-setup is not for ACP sessions. Plugin marketplaces are managed from /plugins or qwen plugins marketplace. Skills, plugins, and workflows can register more commands; inspect them with /commands.

Shell subcommands

The following groups expose these actions. Read their --help output for arguments and scopes, then follow the linked guide.

Group	Actions	Guide
`qwen mcp`	`add`, `add-json`, `remove`, `get`, `list`, `enable`, `disable`, `reset-project-choices`	MCP
`qwen plugins`	`list`, `validate`, `install`, `uninstall`, `enable`, `disable`, `update`, `marketplace`	Plugins
`qwen skills`	`list`, `enable`, `disable`, `install`, `link`, `uninstall`	Skills
`qwen agents`	`list`	Agents
`qwen hooks`	`migrate`	Hooks
`qwen update`	`--check`	Installation
`qwen rollback`	`--to`	Installation

Update and rollback depend on installation state and an available release source. See installation and updates for installation instructions. See headless mode for scripted input and output.