OpenClaw Commands: The Complete Reference
OpenClaw gives you two ways to control your agent: CLI commands (run in your terminal) and slash commands (sent through chat like Telegram). This guide covers both, with clear explanations of what each command does.
CLI Commands (Terminal)
These commands are typed into your terminal to manage the OpenClaw process itself.
Core Commands
openclaw init— Set up a new OpenClaw workspace and run the configuration wizardopenclaw start— Start your OpenClaw agent in the foregroundopenclaw start --background— Start OpenClaw as a background daemonopenclaw stop— Stop the running agentopenclaw restart— Restart the agent (useful after config changes)openclaw status— Check whether the agent is currently runningopenclaw --version— Display the installed version of OpenClawopenclaw help— Show available commands and options
Gateway Commands
The gateway is OpenClaw’s internal routing service — it connects your agent to channels like Telegram.
openclaw gateway start— Start the gateway serviceopenclaw gateway stop— Stop the gateway serviceopenclaw gateway restart— Restart the gatewayopenclaw gateway status— Check gateway health and connection state
Plugin Commands
Plugins extend OpenClaw’s functionality — adding support for new channels, tools, and integrations.
openclaw plugin install <name>— Install a plugin (e.g.,openclaw plugin install telegram)openclaw plugin list— List installed pluginsopenclaw plugin remove <name>— Uninstall a pluginopenclaw plugin update <name>— Update a plugin to the latest version
Update Commands
npm install -g openclaw@latest— Update OpenClaw to the latest versionopenclaw update— Check for and apply available updates (if supported in your version)
Chat Slash Commands
These commands are sent as messages directly to your agent through Telegram (or another chat channel). They start with a forward slash /.
Session & Control
/status— Show current agent status, session info, model, and active settings/reset— Clear the current conversation context and start fresh/stop— Pause or stop an ongoing task/pause— Pause agent activity temporarily/resume— Resume agent activity after a pause
Memory & Context
/memory— Ask the agent to summarize or review what it remembers about you/forget [topic]— Tell the agent to discard specific information from its memory/context— Display the current loaded context and workspace files
Reasoning & Thinking
/reasoning— Toggle extended reasoning mode (the agent thinks through problems more deeply before responding)/think— Ask the agent to reason through a problem step-by-step before answering
Model & Settings
/model— Show or change the current AI model in use/model claude-opus-4— Switch to a specific model by name/settings— View and change agent configuration settings mid-session
Approval & Permissions
When OpenClaw wants to run a potentially sensitive action (like running a shell command), it may ask for approval. These commands let you respond:
/approve allow-once— Approve the action this one time/approve allow-always— Approve and remember permission for future similar actions/approve deny— Deny the action
Agent Tasks & Subagents
/subagent [task]— Spawn a subagent to handle a specific task in the background/agents— List active subagent sessions/yield— Signal the current session to yield to a spawned subagent result
Cron & Scheduling
/cron list— Show all scheduled cron jobs/cron add [schedule] [task]— Add a new scheduled task/cron remove [id]— Remove a scheduled task
Utility
/help— Show available slash commands/ping— Simple connectivity check — agent responds with “pong”/version— Show the running version of OpenClaw/uptime— How long the current session has been running
Workspace File Controls
Beyond commands, many OpenClaw behaviors are controlled by editing files in your workspace folder:
- SOUL.md — Agent personality, tone, and behavioral rules
- USER.md — Your profile, preferences, and context
- AGENTS.md — Operational instructions and startup routines
- MEMORY.md — Long-term memory (curated summaries of important info)
- HEARTBEAT.md — Checklist for periodic agent check-ins
- TOOLS.md — Notes about connected tools, credentials, and APIs
- memory/YYYY-MM-DD.md — Daily activity logs
Editing these files directly is how you “configure” your agent’s behavior in a human-readable way. No JSON or complex settings panels required.
Tips for Power Users
- Use
/reasoningfor complex tasks — it noticeably improves accuracy on multi-step problems - Combine
/cronscheduling with custom prompts to automate daily briefings - Keep
HEARTBEAT.mdshort (5–10 items) to minimize token usage on frequent check-ins - If a task is running too long, send
/stop— the agent will wrap up and report what it’s done so far
Leave a Reply