Release status: Benchlet 1.0 (4) has been uploaded to App Store Connect but is not yet publicly available on the App Store. This policy does not claim App Review approval or public release.
What stays on your iPhone
Benchlet stores shortlisted Hub items, cached model briefs, selected cloud models, model-server names and addresses, Playbook progress, and recent comparison history locally. A comparison can include your task name, prompt, returned outputs, latency, token use, estimated cost, quality review, winner, reason, date, and device conditions.
Credentials
Your Hugging Face inference token and optional model-server keys are stored in the iOS Keychain. Benchlet does not include a shared developer inference key. Do not place credentials in prompts, reports, or screenshots.
Hugging Face Hub requests
Explore and model pages request public repository metadata, cards, and published evaluation information from the Hugging Face Hub. Hub listings and cards are community content and may be changed by their publishers.
Cloud inference
Only after you approve a cloud run, Benchlet sends your fixed prompt, chosen model, and request through Hugging Face to the exact inference provider shown in the app. Your Hugging Face token authorizes the request. Those services may process, retain, and bill for the request under their own terms and policies. No Benchlet-operated server receives the prompt or output in the current candidate.
Apple Intelligence baseline
On a compatible device and OS, the Apple Intelligence baseline uses Apple’s system language model on the iPhone. Benchlet does not claim that an arbitrary Hugging Face repository runs on-device.
Your server
When you choose Your Server, Benchlet sends the prompt directly to the server address you configured. A local address can keep the phone-to-server connection on your network, but Benchlet cannot determine whether software on that server forwards the data elsewhere. Use HTTPS for remote endpoints and whenever a credential is required.
Analytics, tracking, and advertising
The current candidate includes no Benchlet account, advertising network, tracking SDK, third-party analytics SDK, or Benchlet-operated inference service. App Store analytics and TestFlight diagnostics are controlled by Apple and the settings available to users and developers.
Delete your data
Settings includes Delete All Benchlet Data. It removes saved items, cached briefs, Lab selections and history, Playbook progress where included in the submitted build, compute nodes and their Keychain credentials, safety preferences, cloud permission state, the Hugging Face token, and the stored username from that iPhone.
Exports
You control any report you export or share. A report may contain the original prompt, model outputs, model/provider names, measurements, quality ratings, winner, and environment. Review it before sending it to another person or service.
Community content controls
Benchlet hides known mature material by default and provides in-app Hide, Report, and creator-blocking controls. Reports about a public repository may also be sent to chris@teryli.com. Teryli Systems will review the report, update Benchlet safeguards when appropriate, and direct Hub-level removal requests to Hugging Face. Do not include private prompts, outputs, or credentials in a report.
Contact and effective date
This policy is effective July 23, 2026. Questions or privacy requests may be sent to Teryli Systems LLC at chris@teryli.com. Support information is also available on the Benchlet support page.