You send every reply.
Insert puts a draft in your reply box. Review it, make any changes, and use your app’s send button. ToneBird never sends on your behalf.
See what stays on your device, what a model receives, and what only you can approve.

Insert puts a draft in your reply box. Review it, make any changes, and use your app’s send button. ToneBird never sends on your behalf.
Learned adjustments wait for your approval. They only change how ToneBird writes after you review and apply them.
Export or delete your local data in Settings → Privacy & Data.
Your tone profile, person cards, learned corrections, and usage counts are stored locally.
Accessibility helps ToneBird recognize reply fields and read the current conversation when you trigger it.
If text cannot be read, ToneBird tries local text recognition first. A vision model may receive the cropped image if needed for that one read. The image is then discarded.
The context and instructions needed for your reply go to your selected model provider. With a managed model, ToneBird forwards the request. With your own model key, it goes directly to that provider.
Diagnostics stay on your device unless you submit a report. Preview and remove optional conversation excerpts, replies, or screenshots before submitting.
ToneBird does not train foundation models on your conversations. The model provider handles each request under its own policy.
We do not sell your personal information or share it for behavioral advertising across contexts.
Your tone profile, person cards, learned corrections, and usage counts are stored on your device.
Generating a reply may still send the context it needs to an external model. Local storage does not mean all processing happens on your device.
Yes. The model provider receives the context and instructions needed for your reply.
With a managed model, ToneBird forwards the request. With your own model key, writing requests go directly to that provider. See providers and data handling.
When text cannot be read through Accessibility, ToneBird first tries text recognition on your device.
If that is not enough, a vision model may receive the cropped screenshot for that one read. The image is discarded after reading.
No. Learned adjustments wait for your approval.
Review an adjustment and apply it before it changes how ToneBird writes.
ToneBird does not train foundation models on your conversations.
The AI provider processes the content needed for a request under its own policy. Read the provider details.
Open Settings → Privacy & Data in ToneBird to export or delete your local data.
For account or submitted feedback data, contact support@amplift.ai.