How-To Guide

How to Turn Off Reasoning in Hermes Agent and OpenClaw

Q: Is there a switch for reasoning in the dashboard?

No. Use the /reasoning command in Hermes Agent or the /think and /reasoning commands in OpenClaw, typed into the chat.

Q: What is the difference between hiding and turning off reasoning?

Hiding only removes the reasoning text from view. Turning reasoning off asks the model not to think before answering where supported, which is usually faster but can be less careful.

Q: Does this work on OpenClaw?

Yes. OpenClaw uses /think off to disable thinking and /reasoning off to hide reasoning text.

Q: Does the setting survive a restart?

On Hermes Agent, add --global to save it. On OpenClaw, the command applies per session, and a config default makes it permanent.

Q: Will the model do zero thinking with reasoning off?

Not always. Some models and provider routes still reason, so also hide reasoning with /reasoning hide on Hermes Agent or /reasoning off on OpenClaw.

Reasoning models think before they answer, and some chat views show that thinking. Both Hermes Agent and OpenClaw let you hide it or switch it off with a single chat command. There is no dashboard toggle; the command is the switch.

Hide it, or turn it off

Two settings are easy to mix up:

Hide reasoning: the model still thinks, but the thinking text is not shown. Answer quality is unchanged.
Turn reasoning off: the model is asked to skip the thinking step, where it supports that. Replies usually arrive faster and use fewer tokens, but can be less careful on multi-step work such as research or searching an inbox.

Pick hide if the text is only a distraction. Pick off if you want speed or lower cost.

Hermes Agent

Open your bot on the dashboard, click Quick Chat, and send one of these as a message. The same commands work from Telegram and other connected channels.

/reasoning shows the current setting and the options.
/reasoning hide hides the reasoning text.
/reasoning none --global turns thinking off and saves it.
/reasoning medium --global turns it back on at the normal level.

Send each command as its own message, with nothing else in it.

The bot confirms each change, for example Reasoning effort set to 'none' (saved to config). Without --global the effort change applies only to the current conversation. Other levels are minimal, low, high and xhigh, and /reasoning show brings the text back.

On a self-hosted Hermes Agent, the same settings live in ~/.hermes/config.yaml:

agent:
  reasoning_effort: none

display:
  show_reasoning: false

OpenClaw

OpenClaw splits the two settings into two commands. Send them from Quick Chat or any connected channel:

/think off turns thinking off.
/think default returns to the normal thinking level.
/reasoning off hides the reasoning text.
/reasoning on shows it again, as a separate "Thinking" message.

Send each command as its own message. A command sent together with other text applies to that one message only and is not saved.

/think off is stored for the current session until you change or clear it. Reasoning text is hidden by default on OpenClaw, so if you see it, /reasoning off is the fix. Send /reasoning with no argument to see the current value.

On a self-hosted OpenClaw, set permanent defaults in openclaw.json:

{
  "agents": {
    "defaults": {
      "thinkingDefault": "off",
      "reasoningDefault": "off"
    }
  }
}

What to expect

With reasoning off, most models answer directly. Some still produce a little reasoning, and whether a model can fully disable thinking depends on the model and its provider route; on some routes it keeps its own default effort. If you want none of it on screen, turn reasoning off and hide it as well.

Tip: in group chats, keep reasoning display off. Shown reasoning can reveal internal notes to everyone in the group.

Reasoning settings FAQ

Is there a switch for reasoning in the dashboard?

No. Reasoning is controlled by a chat command. Type it into Quick Chat, or into any connected channel such as Telegram, and the bot replies with a confirmation.

What is the difference between hiding and turning off reasoning?

Hiding keeps the model thinking but stops the thinking text from being shown. Turning it off asks the model to skip the thinking step where the model supports that, which usually makes replies faster and cheaper but can make them less careful on multi-step tasks.

Does this work on OpenClaw?

Yes, with different commands. On OpenClaw, /think off turns thinking off and /reasoning off hides the reasoning text. /think default returns to the normal level.

Does the setting survive a restart?

On Hermes Agent, add --global and the choice is saved to the bot configuration. On OpenClaw, /think off is stored for the current session until changed or cleared; a self-hosted install can set a permanent default in the config file.

Will the model do zero thinking with reasoning off?

Not always. Some models still produce some reasoning, and some provider routes keep their own default effort regardless. To make sure none of it is shown, also hide it: /reasoning hide on Hermes Agent, /reasoning off on OpenClaw.