Prompt flags: switch models and modes as you type
Short flags in a prompt switch the model or attach the page.
On this page
A flag is a dash and a letter at the start or end of a prompt, like -p What is this PDF about?. It changes how that
one message is sent: attach the page, think it through, or go to another model. Flags work in the chat and in the
GPT steps of your commands, and HARPA removes them before the prompt
goes to the AI.
# Mode flags
| Flag | What it does |
|---|---|
-p | Attaches the open page to this message, like the Page button |
-t | Thinks the question through before answering, like Thinking in the model menu. On models that offer it, like HARPA AI |
-i | Isolated: sends the prompt without the chat history and system messages. On cloud models and API keys |
-p What is this PDF about? and What is this PDF about? {{page}} send the same thing to the AI. Typing -p lights
up the Page button.
# Model flags
A model flag sends the message to another connection. These come set up:
| Flag | Connection |
|---|---|
-k | Your first API key, see API connections |
-c | Claude in your browser, see Web session connections |
-g | Gemini in your browser |
-x | Grok in your browser |
- With a question, like
-c Rewrite this in plain English, the flag switches the model for that one message. - On its own, like
-cand Enter, it switches the chat to that model.
HARPA's cloud models and ChatGPT have no flag at first: give them one below.
# Combine flags
Write the letters after one dash, or as separate flags:
-pt Is this contract fair to the tenant?
-p -c Summarize this page in five bullet points
In a command, a flag in each GPT step lets one step use a fast model and the next a stronger one:
-k Extract every product name and price from {{page}} as a list.
# Set your own flags
- Open the Models screen: click the model button in the composer, then All connections.
- Click the Advanced settings button next to a section, like Cloud models or Web connections.
- Type a flag in the shortcut field of a connection, like
opusorfast, up to 12 letters, digits or dashes.
Now -opus sends a message to that model. In a team space, only the owner can change the flags of cloud models and
API keys.
# Good to know
- The
-3,-4and-bflags are gone. Give your favorite models their own flags instead. -wand-fare gone too. A prompt that still has one runs as before: HARPA takes any unknown flag out of the prompt. Web search is a tool of the model now: see AI web search.- Automations run on cloud models or API keys: a web session flag in one of their steps is replaced. See Automations.