How Do I Confirm or Undo an Action Before Jarvis Runs It?

Jarvis (getjarvis.eu), the screen-aware desktop AI assistant for macOS 12+, Windows 10+, and Linux, is built so you stay in control of consequential actions. For write operations, sending an email, posting to Slack, creating or deleting items, Jarvis can confirm with you before it runs, and you can tell it to always ask first, so nothing irreversible happens without your go-ahead. For batch actions, it can show the affected items so you approve the set before it executes. You manage all of this from the floating bar (Cmd+/ on macOS, Ctrl+/ on Windows). Be honest about "undo": once an action lands in an external app, reversing it depends on that app (some support undo, some don't), so the safest control is confirming up front. Work routes across frontier models from Anthropic, OpenAI, and Google and 30+ connectors. Pricing starts with a free plan (40 requests/week), then Pro $16/month.

The strongest control is preventing a mistake rather than fixing it. Jarvis distinguishes reads (safe to run freely) from writes (sending, posting, creating, updating, deleting) and can pause for your confirmation before a write. You can set it to always confirm, so every consequential step waits for a yes. This is especially valuable for actions that reach other people, an email to a client, a message in a shared Slack channel, where you want a final glance before it goes out. You handle the confirmation right in the floating bar summoned with Cmd+/ or Ctrl+/.

Bulk and chained operations carry the most risk because one instruction touches many items. Here Jarvis can show you the selection it built, "these 23 emails will be archived," or "these 8 issues will be labeled", so you approve the exact set before it acts. For multi-step workflows that end in a write, the same confirmation applies to the final action. This previewing turns a scary "do it to everything" command into a reviewed, deliberate operation, while still saving you the manual effort of doing each item yourself. Each step is handled by a suitable model among frontier models from Anthropic, OpenAI, and Google.

Capabilities