This page lists commands of OpenClaw 2026.9.8, the self-hosted gateway that connects chat apps to AI agents, as its documentation prints them: the terminal CLI for setup, the Gateway, channels, models, skills and updates, the slash commands typed in a chat, and the terminal UI keys. Rows are worked examples; rarer groups such as fleet, wiki and secrets are left out. Anything starting with openclaw runs in a shell on the Gateway host. Capitals mark values you replace.
Link a local pluginLinks the directory instead of copying it; the path must be on the Gateway host.
openclaw plugins install -l PATH
Install and pin the versionApplies to npm sources: the resolved version is pinned.
openclaw plugins install PACKAGE --pin
Install without enablingRecords the plugin but leaves plugins.allow, plugins.deny and its enabled state alone.
openclaw plugins install PACKAGE --no-enable
Enable pluginsTakes one or more plugin ids in a single call.
openclaw plugins enable codex cua-computer
Disable pluginsDisabling does not uninstall: the plugin's channels stay listed as installed.
openclaw plugins disable IDS
Inspect a plugin--all covers every plugin, and info is an alias of inspect.
openclaw plugins inspect ID
Update all plugins--dry-run previews the update first.
openclaw plugins update --all
Uninstall plugins
openclaw plugins uninstall IDS
Diagnose plugin problems
openclaw plugins doctor
List a marketplace's plugins
openclaw plugins marketplace list SOURCE
List internal hooksEquivalent to the list subcommand; hooks are handlers for command, message, session and Gateway events.
openclaw hooks
List loadable hooksOnly hooks that are enabled, have their requirements met and declare at least one event.
openclaw hooks list --eligible
Check hook eligibility
openclaw hooks check
Enable a hook
openclaw hooks enable NAME
Disable a hook
openclaw hooks disable NAME
Editor and MCP bridges 7
Action
Shortcut
Run the ACP bridgeAn IDE connects to it over stdio and its prompts are forwarded to a Gateway session.
openclaw acp
Attach ACP to a session key--session-label picks an existing session by label instead.
openclaw acp --session agent:main:main
Test the ACP bridgeBuilt-in debug client that spawns the bridge and lets you type prompts without an IDE.
openclaw acp client
Launch Claude Code on a sessionClaude Code gets a temporary MCP grant scoped to a single Gateway session, the main one by default.
openclaw attach
Print the attach configurationWrites the temporary .mcp.json and prints the launch command without starting Claude Code.
openclaw attach --print-config
Serve channels over MCPLets an external MCP client read and send channel conversations through the Gateway.
openclaw mcp serve
Show saved MCP serversShows what is saved without starting any MCP server; the docs suggest it as the starting point.
openclaw mcp status --verbose
Automation and system events 13
The scheduler command registers as openclaw cron with openclaw automations as an alias; every subcommand works with either spelling, and the documentation writes its examples with automations.
Action
Shortcut
Manage scheduled jobsRegistered name of the scheduler command; the docs write its alias, openclaw automations, in every example.
openclaw cron
Create a scheduled jobSchedule first, prompt second; create is an alias of add.
Queue a system eventLands on the main session; --mode now triggers the heartbeat at once rather than at the next tick.
openclaw system event --text "Check for urgent follow-ups" --mode now
Show the last heartbeatenable and disable in its place turn heartbeats back on or pause them.
openclaw system heartbeat last
View presence
openclaw system presence
Diagnostics and repair 20
Action
Shortcut
Run guided health checksInteractive; note that it may migrate legacy config and state even when --fix is not given.
openclaw doctor
Apply recommended repairsAlias of --repair; it still asks before each repair unless running unattended is safe.
openclaw doctor --fix
Read-only lint for CIChanges nothing and sets its exit code from the findings threshold; --severity-min filters them.
openclaw doctor --lint
Doctor report as JSONAdvisory and read-only: it exits successfully, so check ok and findings in the output.
openclaw doctor --json
Scan for extra Gateway installsLooks through system services and reports recent Gateway supervisor restart handoffs.
openclaw doctor --deep
Repair without prompts--non-interactive alone suppresses prompts, not writes; safe migrations still apply.
openclaw doctor --fix --non-interactive
Generate a Gateway tokenCreates and configures a token; restart the Gateway afterwards.
openclaw doctor --generate-gateway-token
Check plugins after an upgradeRuns plugin compatibility probes and exits 1 on any error-level finding.
openclaw doctor --post-upgrade
Hand diagnostics to a coding agentRuns read-only checks, writes a sanitized prompt and offers it to a detected agent such as Claude Code or Codex.
openclaw triage
Collect diagnostics onlyGathers the sanitized diagnostics without starting an agent; --non-interactive does the same.
openclaw triage --json
Audit security settingsStays on the config and filesystem read-only path and does not load plugin runtimes.
openclaw security audit
Audit with live probesAdds best-effort live Gateway probes and plugin-owned audit collectors.
openclaw security audit --deep
Apply safe security fixesSwitches open group policies to allowlist and tightens permissions on state, config and credential files.
openclaw security audit --fix
List sandbox runtimesOne row per runtime with its backend, age, idle time and owning session or agent.
openclaw sandbox list
Recreate sandbox runtimesRemoves them so they are rebuilt with current config on next use; pass exactly one of --all, --session or --agent.
openclaw sandbox recreate --all
Explain the sandbox policyReports the effective mode, scope, workspace access and tool policy, with the config keys to change.
openclaw sandbox explain
Show browser statusPassive: it never launches Chrome just to collect facts.
openclaw browser status
Start the managed browser--headless applies to that start only and does not change the saved profile config.
openclaw browser start
Stop the browser sessionEnds the control session and drops temporary emulation overrides, even for browsers OpenClaw did not launch.
openclaw browser stop
Diagnose the browser--deep adds a live snapshot probe to prove the current tab can be inspected.
openclaw browser doctor --deep
Updates, backup and reset 23
Action
Shortcut
Update OpenClawRestarts the Gateway service after a successful update unless --no-restart is given.
openclaw update
Show update statusInspect availability and the last recorded run before retrying an update.
openclaw update status
Switch update channelChannels are stable, extended-stable, beta and dev; the choice is saved after the core update succeeds.
openclaw update --channel beta
Preview an updateShows the planned channel, target and restart flow without installing or writing config.
openclaw update --dry-run
Update without restarting
openclaw update --no-restart
Skip update confirmationsAnswers prompts such as the downgrade confirmation.
openclaw update --yes
Pick a channel interactivelyAsks for the update channel and whether to restart the Gateway afterwards.
openclaw update wizard
Repair stale update stateFor update status stuck in progress while the Gateway is healthy; make sure no update is still running.
openclaw update repair
Preview recovery cleanupLists retained migration originals once the update and your conversations are verified.
openclaw update cleanup --dry-run
Create a backup archiveWrites a timestamped .tar.gz of state, config, credentials and sessions to the current directory; existing archives are never overwritten.
openclaw backup create
Back up and verifyRescans the new archive the way openclaw backup verify does.
openclaw backup create --verify
Back up the config file onlyExports just the active JSON config, without its $include dependencies.
openclaw backup create --only-config
Verify a backup archiveChecks the manifest, rejects unsafe archive paths and confirms every declared payload exists.
openclaw backup verify ARCHIVE
Restore a backupRestores into a fresh staging directory and leaves the live state directory untouched.
Reset local stateKeeps the CLI installed; scopes are config, config+creds+sessions and full.
openclaw reset
Preview a resetPrints the actions and removes nothing.
openclaw reset --dry-run
Reset everything unattendedfull removes the state directory and workspaces; --non-interactive requires both --scope and --yes.
openclaw reset --scope full --yes --non-interactive
Remove service and local dataWith no scope flags an interactive prompt asks which components to remove; the CLI itself is not removed.
openclaw uninstall
Uninstall everything--all is shorthand for --service --state --workspace --app.
openclaw uninstall --all --yes
List migration providersShows the installed providers; the bundled ones cover Claude, Codex CLI and Hermes.
openclaw migrate list
Preview a Hermes import
openclaw migrate hermes --dry-run
Import state from Hermes
openclaw migrate apply hermes --yes
Terminal UI keys 11
Reply delivery to a chat provider is fixed when the TUI starts: there is no /deliver command or settings toggle, so restart the TUI with --deliver to change it.
Action
Shortcut
Send the messageMessages go to the Gateway, or to the embedded runtime in local mode.
Enter
Insert a newlineStarts a new line in the composer; the message is not sent.
Shift+EnterorCtrl+J
Abort the active runFrom the composer it aborts the run; with a question prompt open it collapses the prompt instead.
Esc
Close a pickerA filter typed into the session picker is cleared first; press again to close.
EscorCtrl+C
Clear input, twice to exitEmpties the composer and prints an exit hint; aborting a run is Esc.
Ctrl+C
Exit the TUI
Ctrl+D
Open the model pickerLists the agent's published models and sets the session override; unavailable ones stay visible with a reason.
Ctrl+L
Open the agent pickerSwitches the TUI to a different agent.
Ctrl+G
Open the session pickerLimited to 50 sessions from the past week; use /session with a key for older ones.
Ctrl+P
Expand or collapse tool outputTool calls appear as cards; this switches them between collapsed and expanded views.
Ctrl+O
Show or hide thinkingChanging the setting reloads the session history.
Ctrl+T
Terminal UI commands 15
Action
Shortcut
Open the terminal UIConnects to the Gateway; when started without --session it resumes the last selected session.
openclaw tui
Run the TUI without a GatewayUses the embedded agent runtime; cannot be combined with --url, --token or --password.
openclaw tui --local
Local TUI shortcutAlias of openclaw tui --local, as is openclaw terminal.
openclaw chat
Connect to a specific GatewayWith --url the TUI ignores stored credentials, so pass --token or --password explicitly.
Open a session with delivery onDelivery of replies to the chat provider is off by default and fixed for the whole TUI session.
openclaw tui --session main --deliver
Switch sessionA bare key such as main is scoped to the current agent; a full agent:other:main key switches agent too.
/session KEY
Switch agent
/agent ID
Show a Gateway summaryReports version, per-channel config summaries and the session inventory, archived sessions included; alias /gwstatus.
/gateway-status
Open TUI settingsPanel with the tool-output and thinking-visibility switches; it cannot change delivery.
/settings
Reopen a pending questionBrings back a question prompt you collapsed with Esc.
/question
Abort the active run
/abort
Quit the TUI/quit does the same; pasted with a trailing newline it does not close the TUI.
/exit
Sign in to a provider locallyLocal mode only: it starts the provider sign-in without leaving the TUI.
/auth
Inspect Chrome extension setupRuns on the TUI host, not the Gateway; install and verify are its other subcommands.
/browser-setup
Run a shell command locallyPut it at the start of a line, as in !openclaw config validate; the TUI asks once per session for permission.
!
Chat commands for sessions and models 21
Directives such as /think, /fast, /verbose, /reasoning, /model and /queue are removed from the message before the model reads it, and most of them persist as session settings when sent alone.
Action
Shortcut
Start a fresh sessionArchives the current session first; a model name may follow.
/new
Reset the session in place/reset soft keeps the transcript and reruns startup.
/reset
Name the sessionLeave the title out to see the present name plus a suggested one.
/name TITLE
Compact the session contextAccepts optional instructions after the command.
/compact
Abort the current run
/stop
Export the session to HTMLOwner-only; writes inside the workspace, and /export is the alias.
/export-session
Show or switch the modelArguments are a model name, default, list or status.
/model
Change model for this sessionConfigured defaults stay as they are; --session is the long form.
/model MODEL -s
Change the agent's default modelOwner or admin only; also switches the current session.
/model MODEL -a
Change the global default modelOwner or admin only; requests an update of agents.defaults.model.
/model MODEL -g
Return to the default modelClears this session's selection without writing any defaults.
/model default -s
List providers or modelsOptional provider, page and limit arguments narrow the list.
/models
Set the thinking leveldefault clears the session override; /thinking and /t are aliases.
/think LEVEL
Show or set fast modeTakes status, auto, on, off or default; the effect depends on the provider.
/fast
Turn on verbose outputAlso off and full; meant for debugging, and risky in group chats.
/verbose on
Show model reasoningAlso off and stream; /reason is the alias.
/reasoning on
Turn on elevated modeAlso off, ask and full; /elev is the alias.
/elevated on
Show plugin trace linesReveals only plugin-owned trace and debug lines for the current session.
/trace on
Set the queue modeModes are steer, followup, collect and interrupt; they decide what a message sent mid-run does.
/queue MODE
Steer the active runAdds guidance to the run in progress; /tell is the alias.
/steer MESSAGE
Sign in to a model providerOwner or admin only; shows the sign-in providers and then the connection methods.
/login
Chat commands for status, tools and skills 20
/help, /commands, /status and /whoami are inline shortcuts: they run at once, even inside a longer message, and only for authorized senders. Ordinary commands must be the only content of the message.
Action
Shortcut
Show the help summary
/help
List all commandsGenerated catalog: what appears depends on config flags, the channel, and installed plugins and skills.
/commands
Show available tools/tools verbose adds short descriptions; results depend on agent, channel, sender and model.
/tools
Show runtime statusIncludes uptime for Gateway and host, the health of plugins, and provider quota.
/status
Show detailed plugin healthRequires commands.plugins: true; lists load errors, quarantines and dependency issues.
/status plugins
Show token usage per replySets the footer under each response; other values are off, full and reset.
/usage tokens
Print a cost summaryLocal summary; in the TUI it covers the session, today and the last 30 days.
/usage cost
Show your sender id/id is the alias; like /help it runs inline for authorized senders.
/whoami
Explain context assemblyTakes list, detail, map or json for different views.
/context
Manage the session goalSub-actions cover status, starting, editing, pausing, resuming, completing, blocking and clearing a goal.
/goal
Ask a side questionUses the session as background but leaves future context and transcript history unchanged; /side is the alias.
/btw QUESTION
Run a named skillGeneric entry point that always works; input for the skill may follow the name.
/skill NAME
Draft a skill from the chatProduces one reviewable skill through Skill Workshop.
/learn
Repeat a promptOwner-only; /loop 5m check deploy status creates a fixed-cadence job, and without an interval the loop paces itself.
/loop INTERVAL PROMPT
List loops in this chat
/loop status
Stop loops in this chatA loop name after stop limits it to matching loops.
/loop stop
Answer an approval promptResolves exec, plugin or OpenClaw change approvals.
/approve ID DECISION
List sub-agent runslog and info are the other subcommands, all for the current session.
/subagents list
List thread-bound agents
/agents
Ask the setup helperWorks from an owner DM; in the TUI it returns you to the setup and repair chat.
/openclaw REQUEST
Chat commands for owners and admins 18
Action
Shortcut
Restart OpenClawEnabled by default through commands.restart: true.
/restart
Update OpenClaw from chatOwner only; uses the configured update channel and reports completion or failure in the same chat.
/update
Show the config from chatOwner-only and off until commands.config: true; get, set and unset are the other subcommands.
/config show
Write a config value from chatValidated before the write and persists across restarts.
/config set channels.whatsapp.responsePrefix="[openclaw]"
Show MCP server configOwner-only; requires commands.mcp: true.
/mcp show
List plugins from chatRequires commands.plugins: true; install, enable and disable are owner-only writes.
/plugins list
Show runtime overridesRuntime-only config overrides that need commands.debug: true; set, unset and reset change them.
/debug show
Set the send policyOwner only; the values are on, off and inherit.
/send on
Manage allowlist entriesText-only; subcommands are list, add and remove.
/allowlist
Set group activation to mentionGroup activation mode is either mention or always.
/activation mention
Run a shell command on the hostRequires commands.bash: true; ! COMMAND is the alias.
/bash COMMAND
Check a background shell jobA session id may follow to pick one job.
!poll
Stop a background shell job
!stop
Turn on text to speechOther arguments include off, status, provider, limit and summary.
/tts on
Manage device pairingBundled plugin command with qr, status, pending, approve, cleanup and notify subcommands.
/pair
Check memory dreamingon and off toggle it, for the owner or a Gateway admin.
/dreaming status
Start a support reportOwner-only; exec approval is requested on each use.
/diagnostics
Detach the conversationThe agent session stays open; only this conversation is detached.
/session unbind
Global flags 10
Action
Shortcut
Use the dev profileState moves to ~/.openclaw-dev and the default Gateway port becomes 19001, with derived ports shifted.
--dev
Use a named profileState lives under ~/.openclaw-NAME; a named profile replaces state and config paths inherited from another one.
--profile NAME
Run inside a containerExecutes the CLI in a running Podman or Docker container; OPENCLAW_CONTAINER supplies the default name.
--container NAME
Override the log levelApplies to both file and console output.
--log-level LEVEL
Disable ANSI colorsSetting NO_COLOR=1 has the same effect.
--no-color
Shorthand for openclaw updateRewrites to the update command, which helps in shells and launcher scripts.
--update
Print the versionAll three spellings show the version number and exit.
-V / --version / -v
Stop option parsingWords after it are still read as commands, as in openclaw -- config get gateway.port, but --help becomes a plain argument.
--
Print JSON outputOn bounded reporting commands stdout carries exactly one JSON document; warnings stay on stderr.
--json
Show a command's helpPlugin-added subcommands change independently, so the built-in help is the authoritative list.
COMMAND --help
No row on this page matches that filter.
Notes
Two surfaces, one agent
OpenClaw splits its typeable controls into two surfaces that never overlap. In-session slash commands (/help, /status, /compact) and the TUI keys steer a live conversation. Everything else on this page belongs to the openclaw terminal CLI, which administers the platform: onboarding, channel accounts, model configuration, backups, and the Gateway process itself. If you are typing into a chat, you want a slash command; if you are typing into a shell on the Gateway host, you want the CLI. Inside the terminal UI, Esc aborts the active run, while Ctrl + C only clears the input and, pressed twice, exits.
Isolating state with global flags
Three of the global flags deal with isolation. --dev moves state under ~/.openclaw-dev and shifts the default Gateway port to 19001, so a development instance can run beside the normal one. --profile NAME generalizes this: state lives under ~/.openclaw-NAME with its own config path, and a named profile replaces the paths inherited from another profile, including those of a running Gateway service. --container NAME runs the whole CLI inside a running Podman or Docker container. Global options go in front of the command; options that belong to a command go after its name.
Operating the Gateway day to day
The operational loop revolves around the Gateway, the process the CLI reaches over WebSocket RPC. openclaw gateway runs it in the foreground, openclaw gateway install turns it into a background service, openclaw gateway status and openclaw status --all report on it, openclaw doctor checks and repairs it, and openclaw backup create archives state, config, credentials and sessions before anything risky. Outbound traffic goes through a single command: openclaw message covers sends and channel actions across Discord, Slack, Telegram, Signal, WhatsApp and the rest. For the in-session command set that pairs with this CLI, see the OpenClaw prompt & command reference.
Questions
What is the difference between OpenClaw slash commands and the openclaw CLI?
Slash commands like /help, /status, and /compact run inside an active chat session and steer the conversation. The openclaw CLI runs in a terminal on the Gateway host and operates the platform itself — setup, channels, models, backups, and the Gateway process. One surface drives a session, the other administers the system.
How do I run several isolated OpenClaw instances on one machine?
Use the global flags. --dev keeps all state under ~/.openclaw-dev and moves the default Gateway port to 19001, while --profile NAME keeps state under ~/.openclaw-NAME with its own config path. --container NAME goes one step further and runs the CLI inside a running Podman or Docker container.
Which openclaw commands should I reach for when something breaks?
Start with openclaw doctor, which pairs health checks with guided repairs. openclaw status --all produces a read-only, pasteable diagnosis that includes the security audit, openclaw health fetches a health snapshot from the running Gateway, and openclaw logs --follow tails the Gateway log. openclaw triage collects sanitized diagnostics and offers them to a coding agent on the same machine.
What does openclaw gateway actually run?
openclaw gateway runs the Gateway in the foreground on a WebSocket port, usually 18789, and refuses to start unless gateway.mode=local is configured. Its subcommands install, start, stop and restart it as a background service (a LaunchAgent on macOS, a systemd user unit on Linux, a Scheduled Task on Windows), and status, health, probe and discover query running gateways over WebSocket RPC.
Which key stops a running reply in the OpenClaw terminal UI?
Esc aborts the active run from the composer, and the commands /stop and /abort do the same. Ctrl + C does not stop it: it clears the input and shows an exit hint, and a second press exits the TUI.
Is openclaw cron the same command as openclaw automations?
Yes. The command registers as openclaw cron with openclaw automations as an alias, and every subcommand works with either spelling. The documentation writes its examples with automations, which is why rows such as openclaw automations list use that form.
Every row on this page was matched to a line in the documentation above on . A binding that could not be found there is not listed.
4 rows from the earlier version of this page were removed in that check: duplicates, and bindings the documentation does not list or lists differently.
Found a wrong or missing key? Tell us and it gets checked against the source. How we verify describes the process and its limits.