Telegram Commands Reference
Complete reference for all Telegram slash commands including /new, /retry, /undo, /title, /branch, /compress, /rollback, /stop, /approve, and /deny.
Overview
Your AI employee responds to a set of slash commands that give you direct control over conversations and actions. These commands are typed directly into the Telegram chat, just like any message. They start with a forward slash and are case-insensitive. You do not need to memorize all of them right away — the most commonly used are /new, /approve, and /deny. The rest are power-user features you can adopt as you get more comfortable.
Conversation Commands
The /new command starts a fresh conversation session, clearing the current context. Use this when switching to a completely different topic. The /retry command regenerates the last response if you were not satisfied with it. The /undo command removes the last exchange (your message and the AI's response) from the conversation history. The /title command lets you set a custom title for the current conversation for easier reference later. These commands help you manage the flow of your conversations without any confusion.
Starting a new conversation
Use /new when switching topics.
Retrying a response
If the response was not what you wanted.
Context Management Commands
The /branch command creates a copy of the current conversation that you can take in a different direction without losing the original thread. The /compress command summarizes the current conversation into a condensed context, freeing up token space while preserving key information. The /rollback command lets you revert the conversation to a previous state. These are especially useful during long, complex sessions where you want to explore different approaches or when the conversation has grown very long.
Use /compress during long research sessions. It keeps the key findings but frees up context space so the AI can process more information.
Action Control Commands
The /approve command confirms a pending action like sending an email, making a call, or processing a payment. The /deny command cancels a pending action. The /stop command immediately halts any ongoing operation. These three commands are your safety controls — they ensure nothing happens without your explicit consent. When your AI employee proposes an action, it will wait for your /approve or /deny before proceeding.
Always review the full content of an action before using /approve. Once approved, actions like sending emails or processing payments cannot be undone.
Approving an email
Confirm sending after reviewing the draft.