Slash commands

A cheatsheet for the OpenClaw slash commands your Teammates support in chat, plus where Associates AI behaves differently.

How commands work in chat

Everything you type as a /… message to a Teammate falls into one of three families:

  • Commands — standalone /… messages handled before the model ever sees them. /new, /reset, /skill, and most of the table below are commands.
  • Directives — /think, /verbose, /reasoning, /elevated, /model, /queue, and a few others. These can be dropped inline in a normal message as a one-off hint, or sent as their own message to change behavior for the rest of the session.
  • Inline shortcuts — /help, /commands, /status, /whoami. These run immediately and strip themselves out, so the rest of your message (if any) still goes to the Teammate.

Note: For directives, where you put them changes what they do. Used inline in a sentence — "run this /think high" — it's a one-off hint for that message only. Sent as its own message — just "/think high" — it persists for the rest of the session.

The cheatsheet

Only commands verified against the platform's running OpenClaw version are listed here. Everything is scoped to text chat — LINE, QQBot, voice, phone, and dock-specific commands aren't part of this list.

Sessions & context

Command What it does Notes
/new [model] Archives the current session and starts a fresh one, optionally on a different model Example: /new sonnet — takes a model name or provider, fuzzy-matched against your allowed list
/reset [soft] Resets session state in place /reset soft keeps the transcript; see the mini-guide below
/compact [instructions] Summarizes the conversation so far to stay under context limits, while keeping the same thread Example: /compact keep the decisions and open questions
/stop Aborts the current run
/name <title> Names the current session Example: /name Q3 pricing research

Model & thinking

Command What it does Notes
/model [name|#|status] Shows or switches the model for this session Only lists models on your allowed list — see Model configuration
/models [provider] [page] Lists available models Same allowed-list restriction as /model. Example: /models anthropic
/think <level> Sets a reasoning-effort level: off, minimal, low, medium, or high (some models support more); /think default clears it Directive — inline vs. standalone changes persistence. Example: /think high
/verbose on|off|full Controls how much of the Teammate's process is shown Directive
/reasoning [on|off|stream] Toggles visible reasoning output Directive
/usage off|tokens|full|reset|cost Shows or resets usage stats for the session
/status Shows session status No cost line on this platform — see below

Discovery

Command What it does Notes
/help Lists available commands Inline shortcut
/commands Lists available commands Inline shortcut
/tools [compact|verbose] Lists tools available to the Teammate
/whoami Shows who you're authenticated as Inline shortcut; alias /id
/context [list|detail|map|json] Inspects the current context window
/tasks Lists tracked tasks

Skills & access

Command What it does Notes
/skill <name> [input] Runs a specific skill by name Example: /skill weekly-report — names are on the Teammate's Skills tab; see Skills
/learn [request] Asks the Teammate to learn something new
/allowlist [list|add|remove] Manages the chat-level allowlist Access control is managed in the dashboard — see the note below

Steering & subagents

Command What it does Notes
/steer <message> Nudges a running task without stopping it Alias /tell
/queue <mode> Sets how mid-run messages are handled: steer, followup, collect, or interrupt; /queue default resets it Directive. Without it, messages sent mid-run steer by default. Example: /queue collect
/focus <target> Binds the current Discord thread or Telegram topic to a session Run it inside the thread or topic you want bound; /unfocus releases it
/unfocus Releases a /focus binding
/subagents list|log|info Lists or inspects subagents
/agents Subagent-related command See the official reference for exact behavior

Voice

Command What it does Notes
/tts on|off Turns spoken replies on or off Works out of the box, no setup — see Voice & text-to-speech for the full set of voice commands
/tts audio <text> Sends a one-off audio reply without turning TTS on Example: /tts audio the deploy finished cleanly

Warning: Don't try to manage who can reach a Teammate from inside a chat message. Chat-level /allowlist is limited on this platform — access control lives in the dashboard; see Messaging channels.

What's different on Associates AI

OpenClaw is the engine we run and extend, not a set of YAML files you're expected to hand-edit — the dashboard is the config surface, with review-and-approve built in, so a handful of admin commands are intentionally turned off rather than left half-working:

  • /config, /mcp, /plugins, /bash, and /debug are disabled. These would let a chat message change what a Teammate can do or read — the dashboard's pending-changes-and-deploy flow exists precisely so those changes get reviewed first. This isn't a missing feature; it's the same trust boundary described in Secrets & egress applied to configuration itself.
  • /restart works as documented — it restarts the Teammate's gateway.
  • /model and /models only show your allowed-models list, not every model OpenClaw supports. If a model you expect is missing, it's not on your company's or agent server's allowed list yet — see Model configuration to add it, then deploy.
  • /status doesn't show a cost line. Usage and spend live on the dashboard's Usage page, which is the source of truth for billing — see Team & billing.
  • On Slack, mention the Teammate instead of reaching for a Slack slash command. Each Teammate's Slack app registers its own slash command (/aai by default, configurable per Teammate), but a Slack slash command only ever reaches one app — so with more than one Teammate in the workspace, /aai can't be aimed at a specific one. Mentioning the Teammate first — @Teammate /status — always targets that Teammate and works for every command on this page, which makes it the recommended way to send commands on Slack. It also sidesteps Slack reserving /status for its own use. See Messaging channels.
  • OpenClaw's built-in memory plugin isn't running here. It's replaced end to end by the platform's own memory system, so commands and behavior tied to OpenClaw's stock memory don't apply. See Memory for how memory actually works on this platform.

For the complete reference

This page covers the commands and behavior customers actually reach for day to day, plus everywhere this platform's behavior diverges from stock OpenClaw. For the full command reference — every option, every flag — see the official docs: docs.openclaw.ai/tools/slash-commands. The platform runs OpenClaw 2026.5.x, so that reference applies directly, aside from the differences called out above.

/reset vs. /new vs. /compact

These three get confused constantly because they all "clean things up," but they do different things:

  • /reset wipes the session's state in place — you're starting over in the same conversation. /reset soft does the same thing but keeps the visible transcript, so you can still scroll back and see what was said even though the Teammate's working state is cleared.
  • /new archives the current session entirely and starts a brand new one — optionally on a different model. Use this when you're switching contexts completely, not just clearing clutter.
  • /compact doesn't wipe anything. It summarizes the conversation so far to free up room under the context limit, while keeping you in the same thread with the same history available to refer back to.

The mistake people make most often: reaching for /reset when a long conversation is just getting unwieldy. If you still want the Teammate to remember what you've been discussing, /compact is almost always the right command — /reset is for when you genuinely want a clean slate.

Build your team.

14 days free. No credit card required.

Start Free Trial