Information Capper accesses on your Mac
Capper uses macOS Accessibility APIs to identify the focused application, window, editor, selection, and bounded surrounding text needed for a request. It does not require Screen Recording. Some supported canvas and Electron editors use a guarded, temporary pasteboard fallback; Capper attempts to restore existing pasteboard items and visibly reports a restoration failure.
Local model requests
The Bionic/LM Studio and Ollama presets use loopback addresses on your Mac. Capper does not send those requests to Capper Cloud. If you choose a non-loopback Custom server, your instruction and captured context are sent to the server you selected.
Capper Cloud requests
Cloud mode sends the instruction and bounded captured text through Capper’s Vercel-hosted proxy to OpenRouter and the selected model provider. Before transport, the app removes the source application name, bundle identifier, and window title. Capper does not store conversation history or log prompt, document, or response text.
Licensing and metering
Polar processes purchases and license-key validation. The Cloud proxy stores a keyed hash—not the raw key—linked to Polar customer, benefit, tier, and status identifiers. Usage receipts contain a request ID, friendly model alias, bounded token counts when provided, optional app version, credit-window data, and processing timestamps. They never contain the prompt or response.
Website and support
Vercel Analytics is the only approved website analytics service. If you email support@getcapper.app, the information in your message is used to answer the request. The website and service also rely on Vercel, Postgres/Neon, Upstash, GitHub, Polar, OpenRouter, and the selected model provider.