Skip to main content

​
Slash Command Reference

Slash commands and directives give you direct control over your AI agent’s behavior. Send them as a message on iMessage, Telegram, WhatsApp, Discord, or WebChat — they work identically across all channels.

​
Commands vs. Directives

There are two related systems:
TypeWhat it doesExample
CommandA standalone message handled entirely by the gateway — never reaches the AI/status sent alone
DirectiveA modifier that persists to your session and is stripped before the AI sees your message/model opus sent alone, or :model opus inline
Directives can be sent as standalone messages (they persist to the session) or inline in a normal message (they apply to that message only and don’t persist). 
# Standalone directive — persists to session:
/think high

# Inline directive — applies to this message only:
:think high :model opus
What are the tax implications of converting my LLC to an S-Corp?

​
Session Control

Start a new conversation session. Your memory and workspace files carry over — only the conversation context resets. Optionally specify a model to switch to.
/new
/new opus
/new gpt
/new gemini
Alias: /reset works the same as /new.
Summarize the current conversation context into a condensed form, freeing up tokens. Optionally pass instructions to guide what the summary should focus on.
/compact
/compact focus on the marketing plan discussion
Useful when you’re deep into a long conversation and want to keep going without starting over.
Immediately abort whatever your agent is currently doing and clear any queued messages.
/stop
View or configure how your agent handles queued messages. Supports debounce, cap, and drop options.
/queue
/queue debounce:2s
/queue cap:25 drop:summarize
Send /queue alone to see your current queue settings.

​
Model & Thinking

Switch to a different AI model. This is a directive — it persists to your session when sent alone, or applies to a single message when used inline as :model.
/model
/model list
/model opus
/model gpt
/model gemini
/model openai/gpt-4o
Alias: /models. Send /model or /model list to see a numbered model picker. Use /model <number> to select from the list.
Control how much reasoning your agent does before responding. Higher levels produce more thoughtful answers but use more tokens.
LevelDescription
offNo extended reasoning — fastest responses
minimalLight reasoning for simple tasks
lowBrief analysis before responding
mediumBalanced reasoning (default for most models)
highDeep analysis and step-by-step thinking
xhighMaximum reasoning depth for complex problems
adaptiveAgent decides the appropriate level per message
/think high
/think adaptive
/think off
Aliases: /thinking, /t. Also works inline: :think high.
Control whether your agent sends its reasoning as a separate message before the final response. Useful for seeing how your agent arrived at an answer.
/reasoning on
/reasoning off
/reasoning stream
stream mode shows reasoning as a Telegram draft in real time (Telegram only). Alias: /reason.
Use /reasoning for personal debugging only — avoid enabling it in group chats where internal reasoning output could be visible to others unintentionally.

​
Output & Display

Control voice message output. When enabled, your agent sends voice messages in addition to (or instead of) text.
OptionDescription
offNo voice messages
alwaysVoice on every response
inboundVoice only when you send a voice message
taggedVoice only when response is tagged for TTS
statusShow current TTS settings
providerShow or set the TTS provider
limitSet a character limit for TTS responses
summaryVoice summary only (not full response)
audioAudio-only mode (no text)
/tts always
/tts off
/tts inbound
/tts status
Enable detailed output including tool call results and failure details. Intended for debugging and extra visibility.
/verbose on
/verbose full
/verbose off
full includes all tool output. on shows summaries. Alias: /v. Also works inline: :verbose.
Keep /verbose off in normal use and in group chats — it can expose internal tool output you may not intend to share.
Control whether your agent sends its response after processing.
/send off
/send on
/send inherit
When off, your agent processes your request but doesn’t deliver the response — useful for testing. Also works inline: :send off.

​
Information

Check your agent’s current state — which model is active, context window usage, queued messages, and runtime mode. Also shows provider quota when available.
/status
View detailed token usage for your current conversation.
/context
/context list
/context detail
/context json
ModeDescription
(default)Summary of token usage
listItemized breakdown by message
detailPer-file, per-tool, per-skill, and system prompt sizes
jsonMachine-readable JSON output
Check your sender identity as your agent sees it. Useful when you have multiple channels connected.
/whoami
Alias: /id.
Display a quick reference of the most common commands, grouped by category.
/help
List all available commands for your agent, including any installed skill commands.
/commands

​
Skills & Export

Invoke a specific skill directly by name. Useful when native command limits prevent per-skill commands from being registered.
/skill weekly-report
/skill summarize-emails
See Skills for more on creating and managing skills.
Export your current session to a self-contained HTML file with the full conversation and system prompt.
/export-session
/export-session /path/to/output.html
Alias: /export. Useful for archiving conversations or sharing session context.

​
Inline Directives

Directives work inline in a normal message to modify how your agent handles that specific message only. They don’t persist to the session when used inline.
DirectiveEffect
:think highApply high reasoning to this message only
:model opusUse a specific model for this message only
:send offProcess but don’t deliver the response
:verboseReturn a more detailed response than usual
Example: 
:think high :model opus
What are the tax implications of converting my LLC to an S-Corp?
Directives are stripped from your message before your agent reads it — they only affect processing behavior.

​
Stop Phrases

Need your agent to stop immediately? Just say “stop” in any form:
  • “stop”
  • “please stop”
  • “STOP!”
  • “cancel”
Stop phrases are recognized in 10+ languages — your agent understands urgency regardless of which language you use.
Stop phrases abort the current action and clear the queue, similar to /stop. If your agent is mid-task (like drafting a long email), you’ll need to ask it to start again.

​
Quick Reference

Command / DirectiveTypeWhat It Does
/new [model]CommandFresh session (alias: /reset)
/compact [instructions]CommandSummarize context, free tokens
/stopCommandAbort current action + clear queue
/queueDirectiveView or configure message queue
/model [name]DirectiveSwitch AI model (alias: /models)
/think [level]DirectiveSet reasoning depth
/reasoning on|off|streamDirectiveToggle reasoning output
/tts [option]CommandText-to-speech control
/verbose on|full|offDirectiveVerbose output mode
/usage [option]CommandPer-response usage footer
/send on|offDirectiveOverride message delivery
/statusCommandModel, context, queue, runtime mode
/context [mode]CommandToken usage details
/whoamiCommandSender identity (alias: /id)
/helpCommandGrouped command reference
/commandsCommandFull command list
/skill [input]CommandRun a skill by name
/export-session [path]CommandExport session to HTML
All commands and directives work identically on iMessage, Telegram, WhatsApp, Discord, and WebChat. There are no channel-specific commands (except /tts stream which is Telegram-only).