This is the part most people want explained slowly. Here it is.
What Corivo actually reads
Corivo's screen-reading mechanism is your operating system's accessibility APIs — macOS Accessibility on Mac, UI Automation on Windows. These are the same channels each OS exposes for screen readers, switch control, and other assistive tech. They surface the structure and text of the app you currently have focused: windows, panes, headings, the document you're editing, the ticket title you're reading, the message you're typing.
No screenshots, ever. Corivo does not capture pixels, run OCR over screenshots, or store screen images on either platform. When an app does not expose useful accessibility text, Corivo records the limited app/window context available and relies on connected tools or your explicit command for more detail.
The desktop app that implements this capture path is open source under Apache-2.0 at https://github.com/Principle-Labs/corivo-desktop, so the local screen-context pipeline is inspectable rather than only described here.
What this gives Corivo:
- The text content of the document, ticket, or thread you're focused on.
- Window and tab titles that name what you're looking at.
- The active app's identity (e.g. "Linear", "Slack", "Notion").
What this does not include:
- Screenshots or image files.
- A continuous video recording.
- Anything from apps you don't have focused. Corivo only reads the active window.
- Anything your operating system itself hides from accessibility — for example, password fields, which both macOS Accessibility and Windows UI Automation exclude from the accessibility tree.
- Anything from apps on your exclusion list (see below).
Controls that ship today
- Per-app exclusion list. Settings → Capture & Privacy lets you skip captures from specific apps by bundle ID (Mac) or executable (Windows). Corivo ships with a sensible default list and you can add your own.
- Pause capture. Pause from the menu bar (Mac) or system tray (Windows) at any time.
- Retention. Captures older than 90 days are dropped automatically by the retention sweep.
On the roadmap
Auto-filtering for sensitive app categories isn't built yet. We'd rather be honest about that than ship a docs page that overpromises. What's planned:
- Auto-skipping for password managers (1Password, Bitwarden, the system keychain on each platform).
- Auto-skipping for banking and finance apps.
- Auto-skipping for health-related apps.
- Respecting incognito / private browsing windows.
Until these ship, the controls available are: the per-app exclusion list above, revoke accessibility permission from your OS settings, or quit Corivo.
Why screen context matters
Almost every "AI gives me a generic answer" moment can be traced to the AI not knowing one of three things:
- What you're looking at right now.
- What changed in your project this week.
- Who the people are.
Screen context covers the first. Connected tools — Notion, Slack, Gmail, Calendar, GitHub, GitLab, Linear, Google Docs, Feishu, Sentry, Intercom, Mem, and more — cover the second and third. Corivo's agent uses both to make "look into this" or "respond to this" mean the actual thing on your screen, instead of forcing you to copy-paste it into a chatbot.
Permission model
When you first launch Corivo, your operating system asks you to grant the permissions Corivo needs:
- Accessibility access — required, primary screen-reading mechanism. macOS Accessibility on Mac, UI Automation on Windows.
- Disk access — required. Corivo's agent executes tasks on your behalf (read files, run commands you authorize), and that needs broad disk access. Full Disk Access on Mac; standard user-folder access plus on-demand UAC prompts on Windows.
You can revoke either at any time from your operating system's privacy settings.