Skip to the page

Reference

Slash commands

Type / in any chat, in Conch or a chat app. Conch's own commands act on the app and never reach the model.

This chat

/newIn chat apps too
Start a new chat.
/clearAlso: /reset · In chat apps too
Start afresh in this chat: the model forgets what was said, you keep every message, and Undo puts it back.
/compact [what to keep]Also: /summarise, /summarize, /compress · In chat apps too
Summarise the start of a long chat, so the model reads less.
/goal [goal | clear]Also: /objective · In chat apps too
Say what this chat is for: kept in mind in every reply, with any model.
/plan [on | off | what to plan]In chat apps too
Read only until you approve its plan, then it acts.
/retryAlso: /again, /regenerate · In chat apps too
Send your last message again.
/undoAlso: /rewind, /revert · In chat apps too
Put back the files the assistant last changed.
/task <what to do>Also: /background, /bg, /delegate
Run it as a task while you keep chatting.
/rename <name>Also: /title
Give this chat a name.
/copy
Copy the last reply.
/exportAlso: /save, /download
Save this chat as a Markdown file.
/resumeAlso: /chats, /history, /continue
Find another chat to carry on.

How it answers

/model [name or words]In chat apps too
Choose the model.
/effort [auto…max]Also: /think, /thinking · In chat apps too
How hard the model thinks.
/fast [on | off]In chat apps too
Turn fast mode on or off.
/agent [name]Also: /agents, /who · In chat apps too
Choose who answers: one of your agents, from the next message on.
/modeAlso: /trust, /permissions, /approvals · In chat apps too
How much it can do without asking.
/folderAlso: /cwd, /cd, /dir, /add-dir
Choose the folder it works in.
/statusAlso: /info, /context · In chat apps too
What this chat is using: model, thinking, mode, folder and goal.
/review [what to look at]
Review the changes in the working folder.
/init
Write an AGENTS.md that explains this project to any assistant.

Conch

/remember <something about you>
Save something to memory.
/memory
See what Conch remembers.
/routinesAlso: /schedule, /cron
Things Conch does on a schedule.
/skills
Things Conch knows how to do.
/commands
Create your own commands.
/appsAlso: /mcp, /integrations, /connect, /tools
Connect apps the assistant can use, with any model.
/providersAlso: /login, /auth, /logout, /keys
Connect or sign in to the models you use.
/usageAlso: /cost, /limits, /stats
See how much usage you have left.
/doctorAlso: /health, /repair, /fix
Check everything and fix what’s broken.
/settingsAlso: /config, /preferences · In chat apps too
Open settings.
/theme
Switch between light and dark.
/helpAlso: /? · In chat apps too
What can I do here?.

Finding one

Type / at the start of the message box. A list opens above it, grouped by what each command acts on: this chat, how it answers, and Conch.

Keep typing to narrow it. A few letters are enough (/cpt finds /compact), and other names work too (/reset finds /clear).

  • ↑ and ↓ move, ⇥ or ↵ chooses, Esc closes.
  • A command with choices shows them in the same list. /effort shows the model's thinking levels, /model every model from every provider you connected, and /mode, /fast and /theme their own. The one in use is ticked and says Current. The arrow at the top goes back.
  • /model with words chooses the model they mean: /model opus 5.5, /model cheaper, /model my Claude plan.
  • /goal and /plan take your own words. The list only suggests (Clear the goal, Plan first); ↵ sends what you typed.
  • On a phone the list sits above the keyboard, with rows big enough for a thumb and ✕ to close it.

Which command wins

A name can mean four things. Conch tries them in this order:

  1. One of Conch's own commands, above.
  2. One of your commands: a prompt you saved in Skills → Your commands, where {{input}} is replaced by what you type after the name.
  3. One of your skills, by name.
  4. One of the provider's own commands, sent as it is.

Conch's newer commands (/goal, /plan, /retry, /export and the others added since) give way to a command or skill of yours with the same name. Nothing you made stops working.

Where a provider has a command of the same name, Conch's is the one listed. It hands over to the provider's when that does the job better: /compact, /review and /init with Claude Code or Codex.

Every provider

Conch's commands work the same whichever model answers, because Conch does them itself.

  • /clear starts the chat afresh. From that line on, your assistant reads nothing above it, and the provider starts a new session. Every message stays for you, and so do the goal, what the chat is held to and what it has read. Undo, on the line and in the note, puts it back until you send something. /new starts a new chat instead.
  • /goal says what the chat is for, like /goal get the release notes for 2.4 written. Your assistant keeps it in mind in every reply, through /clear and whichever model answers. It shows as one line above the message box: press it for Edit or Clear goal. /goal alone shows it, and /goal clear takes it away. In a new chat, it goes with your first message.
  • /agent chooses which of your agents answers, from your next message on. /agent alone lists them, the one answering ticked. /agent atlas chooses by name. The new agent reads the whole chat and answers in its own voice.
  • /plan turns Read only on or off. Your assistant reads and plans, changes nothing, and shows its plan with Start and Keep planning. Start ends Read only and begins the work in the mode you had before (Ask first, if your default is Read only). /plan tidy up this folder turns it on and asks in one go. Claude Code shows its own plan. Every other provider that can use Conch's tools does it through Conch.

In chat apps

The same commands work from a chat app: Telegram, Slack, WhatsApp, Signal, iMessage, email and the rest. They come from the same list, mean the same with every provider, and never reach the model as typed.

Anyone you let in

/newIn groups too
Start a fresh conversation.
/clearAlso: /reset
Start afresh: the model forgets what was said, and Undo puts it back.
/compact [what to keep]Also: /summarise, /summarize, /compress
Summarise the start of a long chat, so the model reads less.
/retryAlso: /again, /regenerate
Send your last message again.
/undoAlso: /rewind, /revert
Take back the last /clear.
/stopIn groups too
Stop what it’s doing.
/helpAlso: /? · In groups too
What I can do here.

You, in your private chat

/goal [goal | clear]Also: /objective
Say what this chat is for: kept in mind in every reply, with any model.
/plan [on | off | what to plan]
Read only until you approve its plan, then it acts.
/model [name or words]
Choose the model.
/effort [auto…max]Also: /think, /thinking
How hard the model thinks.
/fast [on | off]
Turn fast mode on or off.
/agent [name]Also: /agents, /who
Choose who answers here: one of your agents.
/modeAlso: /trust, /permissions, /approvals
How much it can do without asking.
/statusAlso: /info, /context
What this chat is using: model, thinking, mode and goal.
/settingsAlso: /config, /preferences
Choose how Conch works here.
  • Choices are buttons. /model shows the models of the provider you use, the one in use ticked, and Other providers a step away. Tap one and it's done. /effort, /fast and /mode work the same way. Where an app has no buttons (WhatsApp, Signal, iMessage, email, SMS, DingTalk), the choices are numbered: reply with the number.
  • Words after the name work too: /model opus, /model cheaper, /model my Claude plan, /effort high, /fast off, /goal get the 2.4 notes out, /plan tidy up this folder.
  • /clear has an Undo under it, and /undo does the same, until you send something. /undo in a chat app takes back a clear; to put back files, open the chat in Conch.
  • /plan turns Read only on or off at once, with a button to change it back (Plan first or Act as usual). The plan arrives as a message with Start and Keep planning, or reply 1 or 2, yes or no.
  • /agent shows your agents, the one answering ticked, and the others as buttons (numbered where there are none). /agent atlas chooses by name. The choice stays for that chat app. Only you can change it.
  • /goal before your first message goes with it, as in Conch.
  • A mistyped command hears what you probably meant, with a button for it: /modle → /model.
  • Commands only Conch has (/export, /folder, /theme…) say so, and where to find them.
  • Who may use which. Anyone you let in can start afresh, clear, retry and stop in their own chat. What changes how Conch works (agent, model, effort, mode, goal, plan, settings) is yours alone, typed in your private chat. Someone else, a forward or a group can't. In a group, only /new, /stop and /help work.
  • The app's own menu. Telegram lists the commands when you type /, and Discord offers them as its own. Teams lists ten in its menu. Slack keeps / for itself, so there they go after /conch: /conch model.

Thinking effort

/effort takes one of these. A model only offers the levels it has.

/effort autoAuto
The model decides how long to think.
/effort lowLow
Quick answers for simple things.
/effort mediumMedium
A balance of speed and care.
/effort highHigh
Careful thinking for real work.
/effort xhighExtra
Extra thinking for tricky problems.
/effort maxMax
As much thinking as it takes.