Chat Commands

Slash Commands for Your Agent

OpenClaw supports a set of slash commands that control agent behavior. These commands work everywhere your agent receives messages — the Chat tab in your dashboard, WhatsApp, Signal, iMessage, and email. They are processed by the OpenClaw gateway, not by the AI model, so they always work regardless of which model you have configured.

Commands work with or without a space after the slash. /status and / status both work.

Command Reference

Command	What It Does
/status	Shows current session info: active model, token count, and estimated cost.
/new or /reset	Starts a fresh conversation. Clears the current session context so the agent forgets previous messages.
/compact	Compresses the conversation context into a summary. Useful when you are approaching token limits on a long conversation.
/think off|minimal|low|medium|high|xhigh	Controls how much the agent reasons before responding. Higher levels produce more thorough answers but take longer.
/verbose on|off	Toggles verbose output. When on, the agent shows tool calls and intermediate reasoning steps in its responses.
/usage off|tokens|full	Controls the usage footer on each response. tokens shows the token count, full adds a cost estimate.
/model <provider/model>	Switches the AI model mid-conversation. See examples below.
/activation mention|always	For group chats only. mention means the agent only responds when @-mentioned. always means it responds to every message.
/restart	Restarts the gateway process. Owner-only command. Useful after configuration changes or to clear a stuck state.

Examples

Check what model you are using:

Send /status and your agent responds with something like:

Model: anthropic/claude-opus-4-6 | Tokens: 12,450 | Cost: $0.24

Switch to a faster model for quick tasks:

/model anthropic/claude-haiku-4-5

The agent confirms the switch and uses the new model for all subsequent messages in the session.

Switch back to a more capable model:

/model anthropic/claude-sonnet-4-6

Reset after a long conversation:

Send /new and the agent starts fresh. Previous messages are no longer in context. This is useful when you are switching topics and do not want old context influencing responses.

Save tokens on a long thread:

Send /compact and the agent compresses the entire conversation into a summary. It continues responding with full context but uses significantly fewer tokens. This can save money on long-running sessions.

Get more detailed responses:

/think high

The agent takes more time to reason through its answer before responding. Useful for complex questions, code review, or analysis tasks.

See what tools the agent is using:

/verbose on

Responses now include details about which tools the agent invoked, what it searched for, and how it arrived at its answer. Send /verbose off to return to clean responses.

Track costs in real time:

/usage full

Every response now includes a footer with the token count and estimated cost. Helpful for staying aware of API spend during heavy usage.

Tips

• Commands are instant. They do not consume tokens or cost anything — the gateway handles them before they reach the AI model.
• Model names follow the provider/model format. For example: anthropic/claude-sonnet-4-6, openai/gpt-4o, google/gemini-2.5-pro. If you only have one provider configured, you can often omit the provider prefix.
• /compact is your friend. If your agent starts giving shorter or less coherent responses, it may be running into the context window limit. Use /compact to compress and continue.
• /new does not affect channel connections. Starting a fresh conversation only clears the chat context. Your WhatsApp, Signal, and other channel connections remain active.

Related Docs

• Chat Tab — using the chat interface in your dashboard
• What Is OpenClaw? — background on the agent runtime and gateway architecture
• Console Overview — for advanced configuration beyond chat commands