ScreenCaptureKit, or a file you import
macOS asks for screen-recording permission the first time you capture. If you deny it, import an image instead. There is no hidden display grab and no iOS companion in 1.0.
Mockup to SwiftUI, React, CSV, Mermaid, LaTeX
Point the HUD at a UI, a table, a diagram, an equation, or an error and pick the job. The pixels go to the vision provider you configured — unless you chose Ollama on this Mac.
Keys and emails in the result get scrubbed
After the model replies, a local redactor masks common API keys, tokens, emails, and IPv4 addresses in the text you keep. That does not stop the image from reaching the provider you chose.
ScreenCaptureKit on macOS 14+ for a live crop, plus image-file import when recording is denied or you already have a PNG.
SwiftUI, React/Tailwind, design tokens, CSV/SQLite tables, compiler-error explainer, Mermaid, and LaTeX.
OpenAI, Anthropic, OpenRouter, Gemini, xAI, GitHub Models, or Ollama. Paste a key in Settings. There is no Ohm vision cloud.
One-time purchase. Inference cost is yours, with the provider you already use.
This is bring-your-own-key vision, not on-device OCR. Silo never sees the image. Local Ollama is the path that stays on the Mac.