The value is in when it gets called, not in what it does: add a rule telling the assistant to ask before ending a response, and long unattended runs turn into checkpoints where you redirect the work instead of reviewing it afterwards. Because the answer can include an image, a screenshot is a valid reply — useful when what is wrong is easier to show than to describe.
A confirmation endpoint. The assistant reports what it just did and waits; a terminal UI opens on your machine, and whatever you type — text or an image — comes back as the tool's result.
- One call that pauses the assistant and asks you what to do next, taking a summary of the work just completed as its argument
- Your reply returned into the conversation, with text and image input both supported
- The prompt raised in the terminal app you already use — Warp, iTerm2, macOS Terminal, Kitty, Alacritty or Hyper
Nothing to authorize; it runs as npx askme-cli. ASKME_CLI_TERMINAL picks the terminal it opens in — warp is the default, with iterm2, terminal, kitty, alacritty and hyper also supported. It expects a terminal it can open on the machine running the server, so it suits a local assistant rather than a remote one, and Windows support is not yet confirmed by the author. MIT.
One command — npx askme-cli