Primary workflow
Ask about selected text
Highlight a line in an article and ask a short question. Corivo answers through the lens of your own product and positioning.
← All features
Feature
AI screen context without screenshots.
Most AI tools cannot tell what this refers to. Corivo reads structured app text and metadata from the active desktop window so the agent starts from the thing you are actually looking at.
The problem
Why AI screen context
Without AI screen context, every command becomes a typing exercise. You quote the email, paste the ticket, describe the document. Half the prompt is restating what is already on the screen, and that is before the AI gives a useful answer.
Copying entire pages into a chat tab to ask a short question is a daily friction. You waste keystrokes restating what is already visible, and the AI still does not know what you meant by this paragraph or this thread.
Generic AI tools have no idea which ticket you are currently looking at. They cannot resolve a phrase like 'this ticket' or 'this paragraph' without the user pasting context every time, which defeats the speed advantage of a chat surface.
Image-based approaches send screenshots to a vision model, which is slow, lossy, and costly. Structured text from accessibility APIs is the cleaner path for AI screen context — lower latency, lower cost, less data leaving your machine.
What it does
The useful parts, not another chat surface.
Resolves short commands
This ticket, this paragraph, and this email refer to the active work surface instead of an empty prompt.
Uses accessibility text
Corivo reads structured text exposed by operating system accessibility APIs. It does not need screenshots, OCR, or video.
Combines with tool context
When the active app exposes only a title or partial text, connected tools supply the deeper source of truth.
How it works
What gets captured
Corivo records structured text from the focused app plus foreground app and window metadata, then stores that context locally for search and agent use.
- 01No screen images are stored.
- 02Excluded apps and paused capture stay out of the context stream.
- 03Password fields and protected controls remain hidden by the operating system itself.
Corivo reads from operating system accessibility APIs — macOS Accessibility on Mac, UI Automation on Windows. Both expose the visible text and the structure of the focused window. AI screen context starts from this structured tree rather than from pixels.
Capture happens locally and stays locally. Structured text from the focused app, plus minimal foreground app and window metadata, lands in a SQLite database on your computer. There is no upload, no server-side index.
Excluded apps and paused capture stay outside the context stream. Password fields and protected controls remain hidden by the operating system itself, so Corivo never sees them, even when you would like it to.
On both Mac and Windows, the capture pipeline has the same shape. The agent does not care whether the source was AXTextField on macOS or a UI Automation TextPattern on Windows — both surface as structured text to the local AI agent.
Typical workflows
Where this shows up in real work.
Replying to selected text
Highlight a sentence in any article, ask 'how does this apply to us?' AI screen context gives the agent your selection plus your product context, so the answer is specific instead of generic.
Summarizing the active document
Working through a long PRD or contract? Ask 'TL;DR this.' The agent reads the document via accessibility text and returns the key points without taking screenshots or shipping the file anywhere.
Closing a ticket with context
Linear, Jira, or GitHub issue open in the foreground. Ask the agent to draft a resolution. AI screen context already has the ticket body and the visible comments; the draft starts from real text.
Answering questions about a thread
Long email thread on screen. Ask 'what did we agree to?' The agent reads the thread structure through accessibility APIs and surfaces the decision, with the quote ready to use.
Pulling action items from notes
Meeting notes are open. Ask the agent to extract action items and assignees. The output lands ready to paste into your task tool, generated from what is on screen, not from an LLM guess.
Use cases on video
Watch the capability show up in real workflows.
Related workflow
Email tasks, already done
When an email asks you to write something, Corivo understands the request, drafts the document from your work context, and replies to the thread.
Related workflow
Remember what I forgot
Corivo helps recover the specific thing you agreed to yesterday, even when all you remember is the rough scene.
How it differs
How AI screen context differs from screenshot models
Screenshot-based AI sends pixels to a vision model. That is expensive, slower, and loses semantic structure — headings, lists, and links all become rasterized noise that a vision model has to re-parse.
Accessibility-text-based AI screen context gets the structured tree the operating system already maintains. You get the right answer about what is visible without sending images anywhere, and the agent can address specific elements directly.
Screenshots also create a privacy worst-case (your entire screen leaves your machine). Structured text from accessibility APIs limits exposure to what your apps actually expose for accessibility — usually a strict subset, with sensitive fields hidden by the operating system.
Vision-based screen AI
Sends screenshots to a vision model. Higher cost, higher latency, larger privacy surface. Corivo reads structured text instead.
Browser extension assistants
Live inside one browser tab and only see web pages. AI screen context works across native apps too — Mail, Linear, Slack, Notion desktop, IDEs.
ChatGPT desktop apps
Read the active window but only inside a chat surface. Corivo's agent can act on what it sees — draft a reply, edit the doc, file the ticket.
FAQ
Common questions.
What is AI screen context?
AI screen context is the active window's text and metadata, supplied to an AI agent through operating system accessibility APIs. Instead of the user copy-pasting the document or describing the ticket, the agent starts with what is actually visible on screen.
Does Corivo read my screen as images?
No. Corivo reads structured text from operating system accessibility APIs — macOS Accessibility on Mac, UI Automation on Windows. There are no screenshots, no OCR, no video. AI screen context is text-based by design.
What permissions does Corivo need for screen context?
Accessibility permission on Mac, UI Automation access on Windows. Both are one-time grants in operating system settings. Corivo does not request raw screen recording, so it cannot capture pixels even if the code wanted to.
Can I pause screen capture?
Yes. Pause toggles globally from the menu bar (Mac) or system tray (Windows), and you can exclude specific apps so they never enter the capture stream. Excluded apps cover things like password managers and banking apps by default.
Does AI screen context work for native apps, or only browsers?
Both. Any desktop app that exposes accessibility text — Mail, Messages, Linear desktop, Notion desktop, Slack, IDEs, terminals — can supply screen context. Corivo is not limited to web pages or one chat surface.
Does Corivo capture passwords?
No. Password fields and other protected controls are hidden by the operating system itself; they never appear in the accessibility tree. Corivo does not see them, and we cannot opt back in even if we wanted to.
How is AI screen context different on Mac vs Windows?
The underlying API differs — macOS Accessibility on Mac, UI Automation on Windows — but the capture surface is the same. The agent gets structured text and window metadata on both platforms, with the same controls and the same privacy guarantees.
Related paths
Keep following the thread.
Screen context docs
Plain-English details on accessibility text and controls.
PermissionsWhy Corivo asks for Accessibility and Full Disk Access.
Privacy controlsWhat Corivo reads, stores, and avoids.
Local AI agentHow the agent uses screen context to act.
Work memoryWhat happens to captured screen text over time.
Context awareness privacyDetailed privacy notes for screen capture.