Parse

Drag a box around it. Find out what it is.

You are looking at something on screen and you are not sure what it is. A cryptic error, a message that reads like a ransom note, a dashboard with six overlapping labels. Press a hotkey, drag a rectangle, read the explanation. Ask a follow-up if the first pass was unsatisfying. That is the whole app.

Windows GPL-3.0 C++ / Qt 6

Four steps, then it gets out of the way.

1. Press the hotkey

Ctrl+Shift+/ by default, falling back to Ctrl+Shift+Alt+P if something else already stole the first one. Or double click the tray icon if your hands are on the mouse.

2. Drag a rectangle

The screen dims across every monitor and the cursor becomes a crosshair. Drag a box around whatever is confusing you and release.

3. Read the answer

A small dark popup appears anchored near what you captured, and tokens start streaming in within a second or two. No window management, no tab switching.

4. Ask a follow-up

Type a question and press Enter. The image stays in context for the whole conversation. Press Esc when you are done and it disappears.

A tray icon and nothing else.

Parse has no main window. No splash screen, no "welcome back" dialog, no preferences pane, no onboarding tour. It lives in the system tray and waits. The entire visible surface of the application is a hotkey, a selection rectangle, and a popup.

Capture goes through BitBlt and encodes to PNG, then streams to a vision-capable model over server-sent events so text appears as it is generated rather than after a long pause. A named-pipe lock keeps exactly one copy running per user, and the app is per-monitor DPI v2 aware so the selection lands where you drew it.

Where your screenshots go.

Parse sends the region you select to a third party. That is how it works, and it is worth being precise about what that means before you install it.

What leaves your machine

The rectangle you drew, as a PNG, goes to OpenAI's chat API along with your question. Nothing outside that rectangle is captured or sent.
It goes directly from your computer to OpenAI using your own API key. Locke Werks operates no server in this path and receives nothing: not your captures, not your questions, not the fact that you ran the app.
Anything visible inside your selection is included, so treat it the way you would treat pasting a screenshot into a chat window. Do not box a password manager.
Your API key is read from the OPENAI_API_KEY environment variable at startup. It is never written to a config file by Parse and never transmitted anywhere except to OpenAI.

What stays on your machine

Parse keeps a local copy of every capture on disk under your user profile, so captures accumulate over time. There is no telemetry, no analytics, and no crash reporting.

Known limitations.

Hardware overlays capture as black

Protected video, some Chromium windows with hardware compositing, and certain fullscreen games. The fix is the Windows Graphics Capture API, which is not implemented yet.

Mixed-DPI monitors

The rectangle-to-pixel math assumes a uniform device pixel ratio. If your laptop is at 150%, your external is at 100%, and your selection crosses the boundary, the capture will be off.

No offline mode

Requires an internet connection and a valid API key. If the API is down, so is the app. There is no local model fallback.

The hotkey is fixed

Changing it means editing AppController::initialize and rebuilding. A preferences UI is deliberately not on the roadmap.

Install.

Windows 10 1809+ or Windows 11, x64. You will need your own OpenAI API key with access to a vision-capable model, set in the OPENAI_API_KEY environment variable before launch.

setx OPENAI_API_KEY sk-...

# open a new shell afterwards, or the old one will not see it