What this guide covers
Local recognition describes where the OCR and visual calculations run, not a promise that every part of a web application is offline. The site, browser code and language models must be loaded, and optional account functions can contact an authentication service. Screen-sharing permission is still required separately. Reading these public guides starts neither capture nor account requests.
Set it up
For a translation task, first obtain a readable OCR result. In settings, choose the source and target languages and try browser-local translation if your browser and language pair support it. Availability can vary by device and may require a browser-managed model download. An unavailable pair is not repaired by selecting a different interface language.
Check the result
Cloud translation is an explicit alternative, not a hidden fallback. Configure your own endpoint and authorize sending recognized text. The endpoint receives text plus source and target language fields and must return a supported translated-text field. Review the provider’s data handling and any cost before enabling the action; do not send confidential text merely to test an unknown endpoint.
Before you rely on it
Tasks, sources, drafts, settings and recognition records live in browser storage, while active capture streams stay in memory. Image matching keeps a reduced local reference pattern. Shared devices and browser profiles therefore matter even without a cloud history feature. Clear or export appropriate data when changing devices, and protect locally stored service tokens as carefully as other credentials.
Troubleshooting
Review each selected action independently. A record can stay local while a Webhook sends the same recognized sentence elsewhere. Stopping cloud translation does not automatically remove an external notification action, and clearing history cannot recall text already delivered to another service. Test with harmless words and confirm both destination and consent before using sensitive content.
Product reference
AI visual scene understanding is not implemented. Browser-local translation may itself use models supplied by the browser, but that does not turn screen recognition into a general AI agent. The product does not control other applications’ mouse or keyboard. Keep these boundaries distinct when assessing whether a workflow is suitable for personal, workplace or regulated information.
Browser-local translation is selected by default and requires a supported browser and language pair. Otherwise, configure a cloud translation endpoint and explicitly authorize sending recognized text. AI visual understanding is not available, and the product does not control other apps' mouse or keyboard.
Further reading
- Your first screen recognition task
- Choose a recognition region that stays reliable
- Text changes, keywords and task actions