Walkie Claudie - One Button iPhone Voice-to-Agent Shortcut
TLDR: Map Action Button to a shortcut. Have the shortcut start with Dictate Text. If dictated text starts with “Hey Claude”, route to Claude. Similar conditionals for other providers. No matches? Copy to clipboard.
My current job is in the peninsula, and sometimes I drive down. These drives create a lot of time for reflection, and for a while I would ask Siri to start a voice memo to capture thoughts. After one too many, “I didn’t get that.” I started tinkering with shortcuts.
It turns out you don’t need Voice Memo to use the iPhone’s on-device voice transcription. There’s a “Dictate Text” Action that outputs plain text. Which led to a lot of exploration on what is (and unfortunately what isn’t) possible with shortcuts. The good bit is that plain text matching works quite well, which combined with the “if” action provides a functional routing mechanism.
- Text that starts with “Hey Claude” routes the message to Claude
- Similarly, “Hey ChatGPT” routes the message to ChatGPT
- Unmatched text I dump out to a Note, and also to the clipboard1
The clipboard part turned out to be my most used option, because it hooks into my Blink-shell-to-Claude-code workflow.
Talk to Claude Code via Dictate Text and Blink.sh
There are lots of blog posts pointing to how one can combine a Mac mini, Tailscale, and tmux2 for development while on your laptop and phone. My contribution here is this — use the dictated text from your clipboard, and paste it into Claude code. The workflow is still a bit janky, but usable. Press the action button, talk, paste, hit send. Then Claude Code does its thing. I’ve been “coding” while walking to the grocery store.
Pleas for Developers
For the developers of Blink.sh (fill in name here.) — why do you force me to “Allow Paste” every time I want to paste something? It’s the one bit of friction in this flow that’s a pain. I get that it’s likely for security. But I really want to be able to turn it off.
For the developers of Willow. I wish you had a Dictate Text Action! I would swap the Apple one out for yours in a heartbeat. I’m a paying customer! I would pay more!
For folks at Google. I want a “Ask Gemini” action that takes text as input. Calendar and email requests are routed to Claude right now only because that Gemini action isn’t available!