# Chat commands Canonical: https://godspeedmissioncontrol.com/guide/chat-commands Chat commands are short words that start with a slash, such as `/new`. They work in Telegram, which only a [server installed with the one-line command](/guide/install-server) has: open the chat with your own bot and send the command as a message. The notebook chat and your own computer have no slash commands, so write what you want in words there ([Things you can say](/guide/things-you-can-say)). Most commands come from Hermes, the program that runs your assistant, and [Hermes' own list](https://hermes-agent.nousresearch.com/docs/reference/slash-commands) describes each one in more detail. ## The word context Starts the getting-to-know-you chat: in about five minutes, Godspeed asks where you live, a little about you and what you want help with first. Send it without a slash, in Telegram or in the notebook chat. ```text context ``` `/context` with a slash is a different command: see [/context](#context). ## Set up your bot You meet these once, while you connect Telegram on the notebook's **Connect Telegram** page ([First setup](/guide/first-setup)). ### /start with a code Makes you the owner of your bot and starts the setup chat. When you open the bot from the page's button or QR code and press **Start**, Telegram sends it for you. ```text /start ``` If you used the bot before, Telegram shows no **Start** button: press **Copy** on the page and send that message. The code works once. A message without it gets "To set me up, send me the message shown on your Godspeed setup page." During setup, anyone but you gets "This assistant belongs to someone else." ### Setup chat buttons The setup chat asks with buttons. Typing a button's label counts as tapping it, and `/start` on its own repeats the question. Where it waits for a tap, anything else gets "Tap one of the buttons above, please." - Which AI: **ChatGPT (sign in with your ChatGPT plan)**, **OpenRouter (one key, many AI models)**, **Claude (Anthropic API key)**, **OpenAI (API key)**, **Google Gemini (API key)**, **Another provider**. - When something fails: **Get a new code**, **Try again**, **Pick another AI**. - Whether this is your first Godspeed: **Yes, it is my first**, **No, bring in the one I have on GitHub**, later **Start fresh instead**. When you send an API key here, the bot deletes your message at once and says "Thanks, I deleted your message." After setup the bot answers only you. ## Conversations ### /new Starts a fresh conversation with an empty history. Your notebook, goals and files stay as they are. ```text /new [name] ``` Example: `/new holiday plans` ### /stop Stops what the assistant is doing right now, including programs it started in the background. ```text /stop ``` ### /retry and /undo `/retry` sends your last message again. `/undo` removes the last exchange, your message and its answer. ```text /retry /undo ``` ### /title, /sessions and /resume `/title` names the current conversation, `/sessions` lists earlier ones in this chat, and `/resume` goes back to one. ```text /title /sessions /resume ``` ### /compress Shortens a long conversation into a summary, so there is room for more. ```text /compress ``` ### /bg, /btw and /queue `/bg` runs a request in the background and answers when it is ready. `/btw` asks a side question without interrupting the current answer. `/queue` sends a request once the current answer is finished. ```text /bg /btw /queue ``` Example: `/btw what time is it in Tokyo?` ## The AI model and usage ### /model Shows or changes the AI model. On its own it opens a picker with buttons, and the model you pick is used for this conversation only. With `--global` it stays for good. It switches only between AI services already set up ([Switch to another AI](/guide/switch-ai)). ```text /model [--global] ``` ### /backupai Connects, shows or removes a backup AI: a second AI that steps in when your main AI fails. A short setup takes over the chat for a few minutes. It offers the AIs of the [setup chat](#setup-chat-buttons), and **Connect another one**, **Remove it**, **Keep it**, **Not now**. It needs no AI, so it works while your main AI is down, and it deletes a message with a key as soon as it has read it. See [Connect a backup AI](/guide/backup-ai). ```text /backupai ``` ### /status and /usage `/status` shows the model, the conversation and a short recap. `/usage` shows how much the conversation used and, where your AI service reports it, what is left on your plan. ```text /status /usage ``` ### /context Shows how much room is left in the conversation and what fills it. `all` adds detail. ```text /context [all] ``` ### /help and /commands `/help` shows the main commands. `/commands` lists every command and every skill, page by page. ```text /help /commands [page] ``` Example: `/commands 2` ## Voice notes ### /voice Sets how this chat answers with voice notes. A chat starts with your voice notes getting a voice note back, with the same answer as text. ```text /voice on /voice off /voice tts /voice status ``` - `/voice on`: your voice notes get a voice note back. - `/voice off`: answers come as text only. - `/voice tts`: every answer is spoken too, also to what you type. - `/voice status`: shows which of these this chat uses, for example "Voice mode: On (voice reply to voice messages)". The voice itself is chosen in the notebook. See [Talk to it with voice notes](/guide/voice-replies). ## Approve a dangerous command ### /approve and /deny Before a program Hermes considers dangerous, such as deleting a folder with everything in it, the assistant asks you: ```text This command is potentially dangerous (recursive delete). Reply "yes" to approve. ``` - `/approve`, or "yes" or "y": run it this once. - `/approve session`: allow it for the rest of this conversation. - `/approve always`: allow it from now on. Use this only for a command you understand. - `/deny`, or "no" or "n": do not run it. This is separate from the steps that wait for your yes, which you answer in words ([Approve, stop or allow a step](/guide/steps)). ## Skills as commands Every skill (also called a recipe: step-by-step instructions for one job) is a command too: its name, with hyphens turned into underscores. Anything after it is your request. A plain sentence reaches the same skill. ```text / [request] ``` Example: `/due my timesheet is due every month` Common ones are `/tracker`, `/due`, `/coach`, `/morning_brief` and `/context_setup`. When a skill has the same name as a Hermes command, the Hermes command wins. `/radar` and `/subs` are short for `/mc_radar` and `/subscription_review`. ## Habit buttons When you track a habit in Telegram, or ask to see its progress, the reply can carry a picture of its progress with buttons: - **I did it**: marks today as done, while today is still open for that habit. - **Resume**: shows instead when the habit is paused, and makes it active again. - **Undo**: appears after **I did it**, and takes that answer back. In German they read **Erledigt**, **Fortsetzen** and **Rückgängig**. They work only in your private chat with your own bot. To correct an older day, tell it the habit and the date in words. The evening habit check at 21:00 has no buttons: answer "yes", "no" or "skip" ([Build a habit](/guide/habits)). ## Commands to leave alone > [!WARNING] > `/update` updates Hermes to its newest version, not the one chosen for your server's release. Update the whole server instead ([Update Godspeed Mission Control](/guide/update)). > [!WARNING] > `/yolo` makes Hermes skip every approval question for dangerous commands. Leave it off. - `/restart` restarts Hermes' Telegram connection, which Godspeed already keeps running. - `/sethome` marks a chat as the place where results arrive. Your private chat already is. ## Unknown commands and the menu A slash word that is neither a Hermes command nor a skill does not reach the assistant. Hermes answers "Unknown command" and suggests `/commands`. To send it as ordinary text, send it again without the slash. Typing `/` opens Telegram's command menu. It holds at most 60 entries, so some commands and skills are missing from it, but they still work when typed. ## In the notebook chat The notebook chat sends every message as ordinary text, so `/new` or `/model` do nothing special there. Use its controls instead: - The two menus under the message box pick the AI model and how long it thinks. - The **+** button attaches a picture, a PDF or a text file. - **Clear conversation** starts a fresh conversation in that chat. Your saved notes, goals and files stay. [Update first](/guide/update) if you installed an older release.