Rosseta geliştiren: Apir

Rosseta translates text that is visible on a web page by capturing a screenshot of a region you select. It does not read the page's HTML: it sends the selected image to a vision-capable AI model that you configure, and shows the detected text and its translation in an in-page overlay.

HOW IT WORKS
1. Press Ctrl+Shift+Y (Command+Shift+Y on macOS) or click Translate button from the popup. Feel free to use your preferred shortcut (can be set from extension keyboard shortcuts setting).
2. Drag a rectangle over the area you want to read.
3. Rosseta captures that region, sends it to your chosen provider, and displays the original text and its translation in a draggable overlay.

Because it works from the rendered image, it can read text that is not selectable — for example text inside images, videos, or canvas-rendered interfaces — as long as the region is visible and the model can read it.

OUTPUT
- Original and translation are shown side by side, split into the visually distinct blocks the model detects (a heading, caption, button, and so on), each tagged with its detected language.
- For non-Latin scripts, entries can include a romanized reading (Hepburn, Revised Romanization, Hanyu Pinyin) when the model provides one.
- An optional short description summarises the captured image in your target language. It can be turned off to save tokens.

SETUP
Rosseta does not include an AI model or an account and does not collect payment. You add an API key for a provider you already use and choose a model. The model must support image input (vision); text-only models will not work.
- 12 Built-in providers.
- Custom: with your own base URL, optional headers, and optional query parameters.
- Multiple keys: store more than one key per provider and rotate between them.
- Model lists are fetched from the provider's API, and you can add models manually.

OTHER FEATURES
- Translation history is saved locally and can be searched or cleared.
- Target language is selectable from a full language list (BCP 47, e.g. en-US).
- Light, dark, and system themes.

PRIVACY
- No account, no analytics, no telemetry, and no developer servers.
- API keys, preferences, and history are stored locally with browser.storage.local.
- The screenshot you select is sent directly from your browser to the AI provider you configured, using your own key.

REQUIREMENTS AND LIMITATIONS
- You must supply your own API key from a supported provider. Any usage costs are billed by that provider, not by Rosseta.
- Works on regular web pages. Browser-internal pages (chrome:// pages, the New Tab page, the Chrome Web Store) and other extensions' pages cannot be captured.
- On DRM-protected video, the captured region may be blank.
- OCR and translation quality depend on the model you choose.

Open source under the MIT License. Also available for Chrome.