VPN07

OpenClaw Commands 2026: Every CLI and Slash Command Explained

March 9, 2026 18 min read Commands OpenClaw Reference

What This Guide Covers: OpenClaw has two layers of commands — the CLI (terminal commands like openclaw onboard) and slash commands (chat commands like /model and /compact). Both layers are essential and often confuse new users. This is the complete, up-to-date reference for both, with real usage examples from the community.

One of the most common frustrations among new OpenClaw users on X.com is not knowing which commands exist or what they do. Unlike a traditional app with a graphical interface, OpenClaw is primarily a CLI and chat-driven system. The CLI gives you control over infrastructure (install, start, stop, configure). The slash commands give your agent on-the-fly instructions through any connected chat app — WhatsApp, Telegram, Discord, iMessage, Slack.

User @marvelgirl_eth captured it well: "I'm also a total non-technical beginner so the CLI is a whole new interface for me but its super addictive — 2am and I'm still going." And @danpeguine added a key detail that many miss: "Proactive AF: cron jobs, reminders, background tasks." To unleash these features, you need to know the commands. Let's go through all of them.

Part 1: Core CLI Commands (Terminal)

These are commands you run in your terminal (macOS Terminal, Linux shell, or Windows PowerShell). They control the OpenClaw system itself.

Installation & Setup Commands

openclaw onboard

The starting point for every new installation

Runs the interactive onboarding wizard. Configures your AI provider (Anthropic/OpenAI/local), sets up authentication, and optionally connects your first chat channel. Add --install-daemon to also install it as a background service. This is the first command every new user should run.

openclaw update

Update to the latest version

OpenClaw releases new versions frequently. This command updates the installation. Many users check for updates daily — user @cnakazawa literally said: "OpenClaw is the first software in ages for which I constantly check for new releases on GitHub."

openclaw uninstall

Fully remove OpenClaw

Cleanly removes OpenClaw, its daemons, and optionally its config/state data. Useful when migrating to a new machine or doing a fresh reinstall to fix configuration issues.

Gateway Commands

The Gateway is the core server process that handles all communication between your chat apps and the AI. These commands control it:

# Check if gateway is running openclaw gateway status # Start gateway in foreground (useful for troubleshooting) openclaw gateway --port 18789 # Restart the gateway service openclaw gateway restart # Probe connectivity to the gateway openclaw gateway probe # Check gateway logs in real time openclaw logs --follow
openclaw gateway status

Shows: Runtime state, RPC probe status, service/daemon status, and the dashboard URL. Look for "Runtime: running" and "RPC probe: ok" for a healthy gateway.

openclaw gateway probe

Actively tests connectivity. Useful when your chat app is not responding — this tells you if the problem is gateway connectivity or something else upstream.

Dashboard & UI Commands

# Open the web dashboard (browser UI) openclaw dashboard # Open in a specific browser openclaw dashboard --browser firefox # Start the TUI (terminal interface) openclaw tui

Dashboard vs Control UI

The Dashboard (openclaw dashboard) opens a full web interface at http://127.0.0.1:18789/ with the Control UI. You can chat with your agent directly in the browser without any channel setup. The TUI (openclaw tui) is a terminal-based interface for keyboard-driven navigation. Both are great for testing before connecting Telegram or WhatsApp.

Status & Diagnostics Commands

# Quick status overview openclaw status # Full status report with all components openclaw status --all # Check provider API usage and quotas openclaw status --usage # Run the health check diagnostics openclaw doctor # Check channel connectivity openclaw channels status --probe # List configured channels openclaw channels list

The openclaw doctor command is the most powerful diagnostic tool. It checks for common configuration errors, validates your API keys, verifies the daemon is running, and flags any blocking issues. Always run it first when something stops working.

Memory & Session Commands

# List stored memories openclaw memory list # Search memories openclaw memory search "keyword" # Delete a specific memory openclaw memory delete [id] # Manage sessions openclaw sessions list openclaw sessions describe [sessionId] # Cron job management openclaw cron status openclaw cron list openclaw cron runs --id [jobId] --limit 20

Model & Provider Commands

# List all available models openclaw models list # Check which model is active openclaw models status # Configure a model (writes to config) openclaw config set agents.defaults.model.primary "openai/gpt-4o" # Check agent configuration openclaw agent status # Send a message from the CLI (for automation) openclaw message send --target +15555550123 --message "Hello from CLI"

Part 2: Slash Commands (Chat Interface)

Slash commands are sent as messages in your chat app (Telegram, WhatsApp, Discord, iMessage, etc.). They start with / and must be sent as standalone messages. They do not go to the AI — they are intercepted by the Gateway and executed directly.

How Slash Commands Work

Commands must be sent as standalone messages (the entire message is a command). There is also a concept of directives (like /think, /model) which can be embedded in a regular message as "inline hints" — but only for authorized senders. Unauthorized senders see inline directives treated as plain text.

Essential Information Commands

/status

Show current system status

Returns: active model, provider usage/quota, active session info, connected channels. This is the first command to run when something feels off. Shows Claude usage percentage when available (e.g., "Claude 80% remaining").

/help

Show available commands

Displays all commands available in the current context. Output depends on your configuration — commands that are disabled (like /bash or /config) will not appear if not enabled.

/whoami

Show your sender identity (alias: /id)

Displays your sender ID as OpenClaw sees it. Useful for debugging permission issues — if your commands are being ignored, check if your sender ID matches what is in the allowlist.

/context

Show context window breakdown

Shows how the current context is distributed: per-file, per-tool, per-skill, and system prompt sizes. Essential for managing token consumption. Use /context detail for a detailed breakdown or /context json for machine-readable output.

Session Management Commands

/reset

Start a fresh session

Clears the current context and starts a new session. Alias: /new. Accepts an optional model hint: /new openai/gpt-4o. The most powerful command for clearing accumulated context and token overhead.

/compact

Summarize and compress context

Compresses the current session into a compact summary without losing key information. Optionally add instructions: /compact focus on the technical decisions only. Dramatically reduces token usage for long sessions.

/stop

Abort current running task

Immediately stops any currently running agent task. Use when the agent is stuck in a loop, taking too long, or heading in the wrong direction. More graceful than force-quitting.

/restart

Restart the OpenClaw gateway

Restarts the entire gateway process from your chat interface without needing to access the terminal. Useful when configuration changes need to take effect.

Model Control Directives

These are directives — they control how the AI processes your request. When sent as a standalone message, they persist for the session. When embedded in a regular message, they apply only to that single message.

/model [name]

Switch active AI model

Examples: /model openai/gpt-4o, /model anthropic/claude-opus-4, /model list (shows picker), /model status (shows current). On Discord, opens an interactive dropdown. The most-used command after setup.

/think [level]

Control AI thinking depth (aliases: /thinking, /t)

Levels depend on the model/provider. Common options: /think low, /think high, /think off. Higher thinking uses more tokens but produces better results for complex problems. Use low/off for simple tasks to save on API costs.

/elevated [mode]

Control execution permissions (alias: /elev)

Options: on, off, ask (prompt before each exec), full (skip approvals entirely). Use carefully — /elevated full allows the agent to run shell commands without confirmation. Default is ask.

/verbose [on|full|off]

Control response verbosity (alias: /v)

Shows additional debugging information and tool output. full includes detailed failure text. Keep off in normal use and especially in group chats — verbose mode can reveal internal reasoning you did not intend to expose.

/reasoning [on|off|stream]

Show AI reasoning process (alias: /reason)

When on, sends a separate message prefixed with "Reasoning:" showing the AI's chain of thought. The stream option only works as a Telegram draft preview. Useful for debugging why the agent made a wrong decision.

Usage Tracking Commands

# Show token count per response /usage tokens # Show full token + estimated cost breakdown /usage full # Print local cost summary /usage cost # Disable usage display /usage off

Skills & Sub-Agents Commands

/skill [name] [input]

Run a named skill. Useful when per-skill native commands are not available. Example: /skill summarize or /skill weather Tokyo. Skills are user-invocable extensions that give the agent new capabilities.

/subagents list

List all running sub-agents. Also: /subagents kill [id], /subagents steer [id] [message], /subagents log [id]. Sub-agents are background worker processes the agent spawns for parallel tasks.

/kill [agentId]

Immediately abort a running sub-agent. No confirmation. Use when a sub-agent is stuck in an infinite loop or heading in the wrong direction. Use /kill all to abort everything.

/steer [message]

Redirect a running sub-agent mid-task. If the agent is heading the wrong direction, steer it without aborting. Example: /steer focus only on the Python files, ignore JavaScript. Alias: /tell.

Channel Routing Commands

# Switch which channel receives replies /dock-telegram (or /dock_telegram) /dock-discord (or /dock_discord) /dock-slack (or /dock_slack) # Discord-specific thread binding /focus [threadId] - bind thread to a session /unfocus - remove thread binding # Export current session as HTML /export-session [optional-path]

Host Shell Commands

Requires Permission

These commands require commands.bash: true in your config AND the sender must be in the tools.elevated allowlist.

# Run a shell command on the host machine (alias: /bash) ! ls -la ~/Documents ! python3 script.py # For long-running commands: !poll - check status/output !stop - stop the running bash job # Approve pending exec requests /approve allow-once /approve allow-always /approve deny

Config & Debug Commands (Owner-Only)

# Requires commands.config: true in config /config show /config get messages.responsePrefix /config set messages.responsePrefix="[openclaw]" /config unset messages.responsePrefix # Requires commands.debug: true in config /debug show /debug set channels.whatsapp.allowFrom=["+1555","+4477"] /debug reset # clear all runtime overrides # Allowlist management (requires config writes enabled) /allowlist list /allowlist add [senderId] /allowlist remove [senderId]

TTS & Voice Commands

/tts off # disable text-to-speech /tts always # speak all responses /tts inbound # speak incoming messages only /tts status # show TTS status /tts provider # show TTS provider info /tts limit # set character limit # Discord-specific (Discord reserves /tts) /voice join # join voice channel /voice leave # leave voice channel /voice status # check voice status

Quick Reference Card

Command Type What It Does
openclaw onboardCLIInitial setup wizard
openclaw gateway statusCLICheck if gateway is running
openclaw doctorCLIHealth check + fix suggestions
openclaw updateCLIUpdate to latest version
openclaw dashboardCLIOpen browser control UI
openclaw logs --followCLIWatch gateway logs in real time
/statusSlashSystem status + provider usage
/model [name]DirectiveSwitch AI model on the fly
/resetSlashStart fresh session
/compactSlashSummarize context to save tokens
/think [level]DirectiveSet AI reasoning depth
/stopSlashAbort current task
/kill [id]SlashAbort a sub-agent
/context detailSlashShow token breakdown
/usage fullSlashShow tokens + cost per response
/elevated fullDirectiveAllow exec without approval
/skill [name]SlashRun a named skill
! [command]Host ShellRun shell command on host

Why a Fast VPN Makes OpenClaw Commands Faster

When you send commands through Telegram, WhatsApp, or Discord, they travel from your phone to the messaging platform's servers, then to your gateway machine, then to the AI API, and back. Every hop adds latency. Users in Asia, South America, or regions with suboptimal routing to Anthropic/OpenAI servers can experience 3–8 second delays per command — which is frustrating when you are trying to steer an agent in real time.

A VPN with optimized routing and gigabit speeds can reduce that API latency significantly. More importantly, running OpenClaw in the cloud (AWS, Hetzner, Oracle Free Tier) with a VPN keeps all traffic encrypted and routed through optimal paths. Whether your agent runs on a $30 Mac Mini at home or a cloud VM, VPN07 ensures the connection is fast and stable.

VPN07 — The 10-Year Trusted Choice

Stable 1000Mbps connections for always-on AI agents worldwide

$1.5/mo
Best Price
1000Mbps
Gigabit Speed
70+
Countries
30-Day
Refund

10 years of proven reliability. 1000Mbps gigabit nodes in 70+ countries. Whether you are sending a /model command from Tokyo or running OpenClaw on a cloud server in Frankfurt, VPN07 keeps your connections fast, private, and unrestricted.

Pro Tip: Use Command Aliases

Most OpenClaw commands have shorter aliases. The most useful ones: /t for /think, /v for /verbose, /elev for /elevated, /reason for /reasoning, /id for /whoami. These save significant time when you are frequently switching modes mid-conversation.

Related Articles

$1.5/mo · 10 Years Trusted
Try VPN07 Free