Slash commands.

Type / in the chat input to open the autocomplete menu. Arrow keys to navigate, Enter/Tab to pick, Esc to dismiss. Or run /help to open a reference modal inside Obsidian.

Unknown commands (e.g. /typo hello) fall through and get sent as a literal user message — useful when you actually want to send text starting with /.

Reference

Command	Args	Description
/clear	—	Start a new chat. Previous conversation stays in history.
/regen · /regenerate	—	Regenerate the last assistant response.
/title	<new name>	Rename the current conversation.
/leader	—	Toggle leader mode — this chat can spawn follower agents. See SWARM.md.
/agent	[name|off]	Bind a named agent to this chat (bare = picker, off clears). See AGENTS.md.
/skill	[name]	Attach a skill to this chat — its full instructions ride every send (bare = list, off detaches). See SKILLS.md.
/copy	—	Copy the last assistant response to clipboard.
/note	[name]	Save the last assistant response as a new note.
/save-all	[name]	Export the entire conversation as one structured markdown note in your vault.
/export	—	Download the current conversation as a .md file to your downloads folder. (new in v1.0)
/paste	—	Paste from system clipboard into the chat input.
/prompt	[name]	Insert a saved prompt from the prompt library (bare = picker; replaces the input).
/model	<query>	Switch the active model via fuzzy match.
/provider	<query>	Switch the active provider via fuzzy match.
/system	<text>	Set your standing system-prompt extension — global, applies to all chats, persists. No arg resets to default.
/stats	—	Show conversation stats (conversations, messages, tokens).
/remember	<fact>	Manually save a durable fact to long-term memory.
/forget	<substring>	Delete the first memory fact matching the substring.
/memory	[open|clear]	List recent facts, or open/clear the memory file.
/recap	—	Summarize this conversation and log it to the journal.
/help	—	Open a modal listing every slash command.

Conversation management

/clear

Start a new chat. Clears the message list, attachments, and image strip. The previous conversation stays in history (click the history icon in the header to find it).

/regen · /regenerate

Regenerate the last assistant response. Drops the trailing assistant message and re-runs the last user prompt for a fresh answer.

/title <new name>

Rename the current conversation. The title shows up in the history dropdown.

/title Q3 planning notes

/leader

Toggle leader mode on the current conversation. A leader chat can spawn follower agents — separate chats that work subtasks in parallel and report back. Run again to turn it off. See SWARM.md.

/agent [name|off]

Bind one of your named agents to this conversation — its persona, model routing, and tool ceilings take over for this chat. Bare /agent opens the picker; /agent editor fuzzy-matches a name; /agent off returns the chat to the default assistant. Switching mid-conversation is fine: the transcript records which agent wrote each turn. See AGENTS.md.

/skill [name]

Attach a skill from your skills folder to this conversation — the skill's full instructions join the system prompt of every subsequent send, in plain and agent mode alike. Bare /skill lists skills with their attachment state; /skill pdf-processing fuzzy-matches and toggles; /skill off detaches all. Attachments persist on the conversation. See SKILLS.md.

/stats

Show a Notice with conversation stats: total conversations, total messages, total tokens used. If the @/slash autocomplete has run this session, the Notice also shows autocomplete request/token usage for the session.

Saving & exporting

/copy

Copy the last assistant response to your clipboard.

/note [name]

Save the last assistant response as a new note in your configured Note save folder.

• Without a name: the basename is derived from the first line of the response
• With a name: uses what you provide
• Opens the note in a new split
• Frontmatter records source: Curtis, provider, model, created
• Embeds any attached images as ![[filename.png]]

/note Q3 Strategy Draft

/save-all [name]

Export the entire current conversation as a single structured markdown note inside your vault. Without a name, the note is titled after the conversation. Each message becomes a section:

# Conversation title

## 🧑 You
*provider / model*

Your message here.

## 🤖 Assistant
*provider / model*

Assistant response here.

Great for archiving a useful conversation as a searchable, linkable note.

/export

Download the current conversation as a .md file to your browser's downloads folder (not into your vault). Uses a clean markdown format with metadata header:

# Conversation title

> Provider: openai (OpenAI)
> Model: gpt-6.1-sol
> Started: 2026-07-22 4:54 PM
> Messages: 12

---

## User
*2026-07-22 4:54 PM*

Your message.

---

## Assistant · gpt-6.1-sol
*2026-07-22 4:54 PM*

Response.

---

Attached images are emitted as ![](vault/path.png) references — they won't render outside your vault, but the paths are preserved.

Input helpers

/paste

Paste from the system clipboard into the chat input, replacing whatever is currently there. Useful when you've copied text from elsewhere and want to ask about it.

/prompt [name]

Insert a saved prompt from the prompt library into the chat input — the text replaces whatever is in the box, ready to edit and send. Prompts are managed in Settings → Curtis.ai → Prompt library.

• Bare /prompt opens a fuzzy picker; its Create new prompt… tail saves a new entry and inserts it immediately.
• /prompt <name> fuzzy-matches a name and inserts it directly; no match gets a Notice listing what's available.

/prompt           ← opens the picker
/prompt translate ← inserts the best-matching saved prompt

The book-marked button in the composer opens the same picker. One shared library across all chats and models.

/model <query>

Switch the active model via fuzzy match. Searches both model name and ID across all enabled providers.

/model sonnet       → Claude Sonnet 4.5
/model gpt-6-luna   → GPT-6 Luna
/model llama        → Llama 3.3 70B (via Groq, Together, etc.)

/provider <query>

Switch the active provider via fuzzy match. Sets the provider's first available model as active.

/provider anthropic
/provider groq
/provider ollama

/system <text>

Set your standing system-prompt extension. This is a global setting — it applies to every conversation and persists across restarts, so treat it as "my standing instructions to Curtis". It layers on top of the active harness profile (Settings → Generation): the profile is the base contract, this is your addition. Without an argument, resets to the default.

/system You are a senior Rust engineer. Always explain ownership and lifetimes.
/system                            ← resets to default

Memory

/remember <fact>

Manually save a durable fact to long-term memory. The fact is added to AI/Curtis Memory.md and injected into every future prompt.

/remember I prefer concise answers without preamble
/remember My main project is "Aurora" — a Rust async runtime

/forget <substring>

Delete the first memory whose content contains the substring. Returns a Notice showing what was removed.

/forget Aurora

/memory

Without an argument: show a Notice summarizing the last 8 facts.

With arguments:

• /memory open — open the memory file in Obsidian
• /memory clear — wipe all facts (asks for confirmation first)

See MEMORY.md for how memory works and how to edit individual facts from the settings UI (new in v1.0).

Recap & journal

/recap

Summarize the current conversation in Curtis's voice — 2–3 terse bullets covering what was worked on, what was decided, and what was left open. Curtis reads the last ~30 turns (first 500 characters each) and the summary is appended to the chat itself, so it lives in the transcript like any other message.

• Needs at least 3 messages in the conversation — otherwise you get "Nothing to recap yet".
• Also reachable from the pane's export menu → Recap conversation.
• With Recap journal enabled (Settings → Curtis.ai → Conversations), each recap is also appended to the journal file — default AI/Curtis Journal.md, path configurable right below the toggle. The journal is plain, append-only markdown: one entry per recap, and Curtis never rewrites it, so hand edits are safe. It syncs and searches like any vault note.

Reference

/help

Open a modal listing every slash command. Use this when you forget the syntax.