On-device inference
After you download and load a model, text generation is performed on the device. There is no default path that sends your chat to a remote model provider for completion. Third-party LLM APIs are not required for answers while using local weights.
When the network is used
Network access is primarily for acquiring model files and normal framework telemetry if you have not disabled it. This marketing site may use standard analytics from its host — keep that separate from the app's on-device inference story.
On-device components
OCR (ML Kit), speech recognition (whisper.rn), and Mobile Actions parsing are all designed to run without cloud LLM calls — structured actions and document text stay local.
Tester signups
If you submit the signup form, we store your email (and optional note) in our database to coordinate builds. We do not use that data to train models.
Questions? Reach out through the channel you received this build from, or leave a note on the download page.