Slash commands
Type / as the first character in the composer to open the command palette - a floating list that filters as you type. It is the fastest way to run chat actions, switch model or persona, manage memory, fetch a URL and more without lifting your hands off the keyboard.
The command palette
The palette opens when the composer text starts with / and narrows to the commands whose names begin with what you type - /re leaves /rename, /regenerate and /remember. Text that goes on with a space or a second slash, such as // note or / aside, doesn’t open it and is sent as an ordinary message. Each row shows the command, its arguments and a one-line description, under the same group headers /help uses (the headers appear once the list spans more than one group).
Once you’ve typed a command’s full name, the palette lists what can follow it as rows of their own. Fixed arguments match from the start of the word: /list models, /list snippets and the rest, /instructions clear, /folder none, the personas after /persona, the tones after /tone, the label colours, the /set parameters, on and off, and the /theme modes. Names match anywhere, exact matches first: your models after /model, your Spaces after /space and your Snippets after /snippet. Where the argument is optional - /model, /persona, /tone, /instructions - the bare command heads the list, so Enter runs it as typed. The highlight returns to the top row as you type, so Enter on a name typed out in full runs that one rather than a longer name that contains it.
- ↑ / ↓ - move the selection through the filtered list. It wraps around at either end.
- Tab - complete the highlighted command without running it.
- Enter - complete a partial command, or run one that is already fully typed. Shift + Enter still inserts a newline.
- Esc - hide the palette for the current draft. It comes back once you delete the leading
/and type it again.
Clicking a row completes it, like Tab. Anything that isn’t a recognised command is sent as an ordinary message, so a prompt that happens to start with / is never swallowed. Commands run even while a reply is streaming; the few that can’t, such as /regenerate, say so. Private Chat has no command palette - a leading / there is just text.
Where the results appear
A command that does something confirms with a short toast (Renamed chat, Switched model). One that fails says why in a Command failed toast - for example No active chat. Start one with /new first. when a command needs an open chat and none is open - and stays in the composer so you can fix it and run it again. Commands that return a list - /list, /tools, /fetch, /stats, a bare /instructions, or a name that matches several things - show it in a panel just above the composer. Close the panel with its ×; sending your next message clears it too.
The commands
Commands are grouped the way /help lists them. <arg> is required, [arg] optional; a command shown without an argument takes none and refuses one rather than dropping it (/regenerate doesn’t take arguments.). Most act on the chat that is open.
On a Space’s page the composer starts a new chat, so there is no chat to act on: commands that need one answer This works on a chat. Open one of this Space’s chats, or send a message here to start one. The memory commands work on that Space, /new starts its chat in the Space and opens it, and /space opens the other Space’s page.
Chat
| Command | What it does |
|---|---|
/new | Start a new chat. It lands in the active Space when there is one (the Space you’re viewing, or one picked with /space). Typed on a Space’s page, it opens the new chat and takes along any Library sources you picked in that page’s composer. |
/clear | Delete every message in the current chat, after a confirmation. A reply still streaming in that chat is stopped first, and a compaction summary goes too (your own instructions stay). Refused while the chat is being compacted. |
/rename <title> | Rename the current chat. The whole rest of the line is the title. |
/pin | Pin the current chat, or unpin it if it’s pinned. |
/label <color|clear> | Give the current chat a colour label: red, amber, green, blue, purple or pink. /label clear removes it. |
/folder <name|none> | Move the current chat into the folder with that name (matched exactly, ignoring case), creating the folder if there is none. /folder none takes the chat out of its folder. |
/archive | Move the current chat to the archive, with an Undo in the toast. |
/delete | Delete the current chat permanently, after a confirmation. |
/fork | Fork the chat into a new copy and open it. Refused while a reply in the chat is streaming or still being prepared. |
/regenerate | Re-stream the last assistant reply. Refused when the last turn isn’t an assistant reply, the reply was imported, the chat has no model, or the chat is still busy or preparing a reply. |
/stop | Stop this chat’s reply, like the Stop button - also one still being prepared - or take its prompt out of the queue. Other chats aren’t touched. |
/copy [N] | Copy the latest assistant reply to the clipboard, or the Nth-latest - /copy 2 is the one before it. Only assistant replies count, and N must be a positive whole number. |
/find [text] | Open the in-chat finder, like Search in chat in the chat header, with your text already in it when you give some. |
/export | Open the Export context dialog for this chat. |
/import | Open the Import context dialog for this chat. |
/stats | Show the chat’s message counts, total tokens and average tokens/sec across replies, the last reply’s numbers, and the model. |
/compact | Summarise older messages to free context - see Context & compaction. Only one chat compacts at a time. |
/private | Open Private Chat. A reply streaming in a regular chat is stopped first. |
Model & persona
| Command | What it does |
|---|---|
/model [name] | Switch the current chat’s model. An exact model id or display name wins; otherwise every model whose id or name contains what you typed matches, and several matches are listed so you can be more specific. When Ollama and an OpenAI-compatible server both have a model with the same id, prefix it with ollama/ or openai/. On its own, /model opens the model picker in the chat header. |
/persona [name] | Apply one of the built-in personas to the current chat, matched the same way by name. /persona none clears it, and /persona on its own shows the current one. |
/tone [name] | Set the current chat’s tone, matched the same way; /tone default clears it. On its own, it shows the tone in effect - the chat’s own, or the default from Settings → General. |
/set <param> <value> | Set one of the chat’s parameters, within the parameters panel’s ranges. Only that one value is saved as the chat’s override - see the list below. |
/set reset | Drop the chat’s own parameter changes, like the panel’s Reset to defaults button (Reset to model defaults for a model with its own defaults). |
/unload | Unload the current chat’s Ollama model from memory; it loads again on your next message. If a reply is still using it, it goes once that reply ends. OpenAI-compatible models can’t be unloaded - their server manages them. |
/set takes one plain number (not 1e2 or 0x10), in the ranges the panel’s own controls allow:
| Parameter | Panel setting | Accepts |
|---|---|---|
temperature | Temperature | 0 to 1 |
top_p | Top-P | 0 to 1 |
top_k | Top-K | a whole number, 0 to 200 |
min_p | Min-P | 0 to 0.5 |
num_ctx | Context Length | one of 4096, 8192, 16384, 32768, 65536, 131072, 262144, 524288, 1048576 |
max_tokens | Max Tokens | a whole number, 128 to 32768 |
repeat_penalty | Repeat Penalty | 0.8 to 2 |
frequency_penalty | Frequency Penalty | −2 to 2 |
presence_penalty | Presence Penalty | −2 to 2 |
num_gpu | GPU Layers | a whole number, 0 or more |
seed | Seed | any whole number |
Listings
| Command | What it lists |
|---|---|
/list models | Every model from both providers, with its provider. |
/list personas | The built-in personas and what each does. |
/list spaces | Your Spaces, marking the active one. |
/list snippets | Your Snippets, with any pinned model and the start of the prompt. |
/list mcp | Your MCP servers, enabled or disabled, and how each connects. |
/list providers | The Ollama and OpenAI-compatible base URLs, and whether an API key is saved. |
/list memories | The memories /remember and /forget work on (see below), each with the short id /forget accepts. |
Prompts
| Command | What it does |
|---|---|
/instructions <text> | Set the current chat’s own instructions - the Additional instructions in the parameters panel. Line breaks are kept. |
/instructions clear | Clear them. |
/instructions | On its own, show the chat’s current instructions. A compaction summary stored with them is kept whichever form you use. |
/snippet <name> | Put a saved Snippet into the composer, matched by title, with its variables filled in and any files saved with it attached. Any blanks left open the fill-in dialog first. Unlike a Snippet’s Run, it stays in the current chat and doesn’t switch model. |
Memory & Spaces
The memory commands work on one list: on a Space’s page, that Space’s; otherwise the open chat’s Space; for a chat outside any Space, the active Space if there is one; otherwise your global memories. See Spaces and Memory.
| Command | What it does |
|---|---|
/remember <fact> | Save a fact to that list. Refused when the Space’s memory is switched off - or, outside a Space, while Settings → Features → Global memories is off. |
/forget <id|query> | Remove a memory by its id (at least the eight characters /list memories shows) or by text it contains. When several match the text, they’re listed so you can re-run with an id. |
/space <name> | Make a Space active, matched by name. /new then starts its chat there, and the memory commands use that Space outside a Space chat. Typed on a Space’s page, it opens the other Space’s page. |
Tools & web
| Command | What it does |
|---|---|
/tools | List the tools exposed by your enabled MCP servers, server by server; a server that can’t be reached shows its error. Built-in tools aren’t listed. |
/web-fetch on|off | Turn Web fetch on or off for every chat - the same switch as in Settings → Tools. |
/fetch <url> | Fetch one page now and show its final URL, title, size and the first 200 characters of its text - a quick way to see what Web fetch would get. It works with the Web fetch switch off and goes through the same safety checks. |
/thinking on|off | Turn the current chat’s Thinking toggle on or off (Ollama models that can reason). |
App
| Command | What it does |
|---|---|
/settings [tab] | Open Settings, optionally on a tab: general, providers, features, tools, library, appearance, mcp, archive, data, security, updates or about. |
/theme light|dark|system | Set the Color mode in Settings → Appearance. The repaint is the confirmation. |
/lock | Lock Loach now, like Ctrl + Shift + L (⌘ + Shift + L on macOS). Needs a PIN or password set up in Settings → Security. |
/help | Open the Slash commands reference, with every command grouped as above. |
Recent commands
Above those groups sits a Recent group: up to four of the commands you ran last that match what you’ve typed, hoisted out of their normal groups. Loach keeps the last six command names in its settings rather than per session, so the group is useful on the first / after a restart instead of only once you have already run a command that launch. A command counts even if it failed - you’re likely to retry it - while text that isn’t a command is never recorded. Argument rows such as /list models stay where they are, and so does a bare command heading its own list, like /model.
/clear and /delete route through a confirmation dialog before they run, so a mistyped command can’t wipe a conversation by accident. The dialog’s confirm button has focus, so Enter confirms and Esc cancels. /archive doesn’t ask - it offers Undo instead.