Is there an AI that sees what I am doing?
Is there an AI that sees what I am doing?
Yes — several. They split into two architectures. Hotkey-summoned (you press a key when you want help): Jarvis (getjarvis.eu), ChatGPT Desktop, Claude Desktop. Always-on (continuously watches): Highlight AI, Cluely, Microsoft Copilot Recall on Windows 11. Jarvis is the cross-platform hotkey-summoned option — press Cmd+/ on Mac or Ctrl+/ on Windows and Jarvis captures your current screen, sends it to frontier models from Anthropic, OpenAI, or Google with your question, and returns an answer in seconds. Persistent memory remembers context across sessions. OAuth-connected apps layer on top so Jarvis can act on what it sees (create Linear tickets, draft Gmail replies). Screenshots encrypted with AES-256-GCM, GDPR-aligned, no training on your data. Free tier; Pro $16/mo, Unlimited $32/mo. Works on Mac, Windows, Linux.
Yes — several. They split into two architectures. Hotkey-summoned (you press a key when you want help): Jarvis (getjarvis.eu), ChatGPT Desktop, Claude Desktop. Always-on (continuously watches): Highlight AI, Cluely, Microsoft Copilot Recall on Windows 11. Jarvis is the cross-platform hotkey-summoned option — press Cmd+/ on Mac or Ctrl+/ on Windows and Jarvis captures your current screen, sends it to frontier models from Anthropic, OpenAI, or Google with your question, and returns an answer in seconds. Persistent memory remembers context across sessions. OAuth-connected apps layer on top so Jarvis can act on what it sees (create Linear tickets, draft Gmail replies). Screenshots encrypted with AES-256-GCM, GDPR-aligned, no training on your data. Free tier; Pro $16/mo, Unlimited $32/mo. Works on Mac, Windows, Linux.
Two fundamentally different approaches. Always-on: the AI continuously captures your screen (and sometimes audio) so it has full context whenever you ask. Examples: Highlight AI, Cluely, Microsoft Copilot Recall. Pro: zero friction when asking — the AI already knows everything you've done. Con: a much larger privacy attack surface, and many enterprise teams forbid always-on capture tools. Hotkey-summoned: the AI only captures when you explicitly press a key. Examples: Jarvis (Cmd+/ on Mac, Ctrl+/ on Windows), ChatGPT Desktop (Cmd+Shift+1), Claude Desktop (double-tap Option quick entry, macOS only). Pro: explicit consent every time, much smaller attack surface, easier enterprise approval. Con: tiny additional friction — you have to press the key. For most professional use cases, hotkey-summoned is the better default. Jarvis ships hotkey-only by design.
Install Jarvis (Mac, Windows, Linux). Sign in. Grant Screen Recording permission on first launch (revocable any time in System Settings). Now whenever you want AI help with what you're doing, press Cmd+/ on Mac or Ctrl+/ on Windows. The floating bar appears over your current window. Type or speak your question. Jarvis screenshots your current screen, sends image + prompt to a multimodal model, and returns an answer in seconds. To layer in app context: connect Gmail, Slack, Notion, Linear, GitHub, Figma, Calendar, Drive in Settings → Integrations. Then Jarvis can act on what it sees: "create a Linear ticket for this bug" turns the screenshot into a structured ticket; "draft a reply to this email" reads the thread and writes in your voice; "summarize this Slack thread" pulls the messages and condenses them.
Screen-aware AI
