Plan locally
Your words become steps
The on-device language model turns your sentence into a plan of small, closed primitives: open this app, press this control, type this text. No cloud round-trip, ever.
Feature
Say the errand.
Watch it run.
Hold ⌃⇧A and speak: “message Priya on Slack that I’m running late,” “search YouTube for the match highlights.” Velora plans the steps on your Mac with the same local models that clean your dictation, then carries them out in the app you’re already in. With the open-source Cua Driver installed, it can also drive another app in a background window while yours stays in front.
How it works
An agent, not a macro.
Recorded macros break the moment a window moves. Action Mode works the way you would: do a few steps, look at the screen, adjust.
Read the screen
Reality checks between steps
Velora reads what the window actually says as it goes: pressing the right person’s chat when a shortcut didn’t land, fixing a misheard name to the spelling it can see in front of it.
Stop, don’t guess
Wrong window, no action
If the window in front isn’t the one it expects, Velora stops rather than act on the wrong thing. A half-done errand you can see beats a confident mistake you can’t.
“Draft”
Written, but unsent
Say “draft” and Action Mode stops with the message composed but not sent. The send stays yours. It also keeps a private, local record of what it did, cleaned up automatically.
The fine print
- Action Mode uses a closed set of primitives — it cannot run arbitrary shell commands, and Velora refuses to add that on purpose.
- It acts through the same accessibility permission as dictation; there is no network port and no remote control surface.
- It is new and deliberately conservative: expect it to stop and tell you rather than push through ambiguity.
- Every action happens on your Mac. Without the optional Cua Driver, Velora only acts in the app in front of you and refuses errands that need another app; with it, “Work in the background” drives the target app’s window while your current app stays in front. You can take the keyboard back at any time.
- Web tasks read the front browser tab’s address: through accessibility in Safari and Firefox, and in Chrome-family browsers through the Automation permission that dictation asks for once. Action Mode never prompts for it and never reads page contents that way.
- With the Cua Driver, when macOS can’t expose a pinned background window’s accessibility tree, Velora reads an exact screenshot of that window using Apple’s on-device text recognition. The image and the text stay on your Mac, are read-only evidence, and never authorise a click on their own.
Free. Open source. Local.
Stop typing. Keep writing.
Already using Velora? A star is how the next Mac user finds it.